row_id string | section string | section_slug string | resource_type string | marker string | title string | url string | url_kind string | domain string | annotation string | description string | key_contribution string | novelty string | impact string | signal string | signal_strength string | source_readme string | source_line int64 | source_url string | date_added string | collection string | collection_slug string | user_goal string | lifecycle_stages string | audience string | loop_layer string | scope_fit string | evidence_class string | evidence_tier string | source_status string | canonical_url string | source_title string | source_description string | authors string | publication_date string | publication_year string | publication_venue string | publisher string | doi string | publication_note string | primary_category string | metadata_source string | github_repo string | github_stars string | github_forks string | github_license string | github_created_at string | github_updated_at string | arxiv_id string | audited_at timestamp[ms] |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ale-0301 | Coding-Agent Loop Systems | coding-agent-loop-systems | Tool | 🧰 | DeerFlow | https://github.com/bytedance/deer-flow | external | github.com | ByteDance's open-source long-horizon SuperAgent harness that researches, codes, and creates across extended multi-step runs, one of the most widely adopted open agent harnesses. | ByteDance's open-source long-horizon SuperAgent harness that researches, codes, and creates across extended multi-step runs, one of the most widely adopted open agent harnesses. | ByteDance's open-source long-horizon SuperAgent harness that researches, codes, and creates across extended multi-step runs, one of the most widely adopted open agent harnesses. | The work targets tasks that exceed a single context window or prompt session. ByteDance's open-source long-horizon SuperAgent harness that researches, codes, and creates across extended multi-step runs, one of the most widely adopted open agent harnesses. | Use DeerFlow to choose an implementation surface for repeatable agent work. | Inspectable GitHub source (78,131 stars; 10,655 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 926 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L926 | 2026-07-22 | Build | build | Choose runtimes, tools, and delegation surfaces. | workspace;delegation;verification;state | builder | agent | direct | source-implementation | A | ok | https://github.com/bytedance/deer-flow | GitHub - bytedance/deer-flow: An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours. · GitHub | An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours. - bytedance/deer-flow | 2025-05-07 | 2025 | bytedance/deer-flow | GitHub | github-api | bytedance/deer-flow | 78131 | 10655 | MIT | 2025-05-07T02:50:19Z | 2026-07-29T08:04:53Z | 2026-07-29T08:06:54 | |||||
ale-0302 | Coding-Agent Loop Systems | coding-agent-loop-systems | Tool | 🧰 | Grok Build | https://github.com/xai-org/grok-build | external | github.com | xAI's official agentic coding CLI that plans, edits, and verifies changes in a repository, joining the major vendor coding-agent runtimes. | xAI's official agentic coding CLI that plans, edits, and verifies changes in a repository, joining the major vendor coding-agent runtimes. | xAI's official agentic coding CLI that plans, edits, and verifies changes in a repository, joining the major vendor coding-agent runtimes. | Primary-source operational guidance rather than commentary. xAI's official agentic coding CLI that plans, edits, and verifies changes in a repository, joining the major vendor coding-agent runtimes. | Use Grok Build to choose an implementation surface for repeatable agent work. | Inspectable GitHub source (23,323 stars; 4,423 forks; Apache-2.0 license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 927 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L927 | 2026-07-22 | Build | build | Choose runtimes, tools, and delegation surfaces. | verification | builder | agent | direct | source-implementation | A | ok | https://github.com/xai-org/grok-build | GitHub - xai-org/grok-build: SpaceXAI's coding agent harness and TUI. Fullscreen, mouse interactive, extensible. · GitHub | SpaceXAI's coding agent harness and TUI. Fullscreen, mouse interactive, extensible. - xai-org/grok-build | 2026-07-14 | 2026 | xai-org/grok-build | GitHub | github-api | xai-org/grok-build | 23323 | 4423 | Apache-2.0 | 2026-07-14T20:04:23Z | 2026-07-29T08:00:07Z | 2026-07-29T08:06:54 | |||||
ale-0303 | Coding-Agent Loop Systems | coding-agent-loop-systems | Tool | 🧰 | BlitzOS | https://github.com/blitzdotdev/blitzos | external | github.com | Open-source pattern for booting cloud agents pre-loaded with your work context so recurring sessions resume instantly instead of cold-starting. | Open-source pattern for booting cloud agents pre-loaded with your work context so recurring sessions resume instantly instead of cold-starting. | Open-source pattern for booting cloud agents pre-loaded with your work context so recurring sessions resume instantly instead of cold-starting. | Context is managed as durable loop state rather than a single prompt payload. Open-source pattern for booting cloud agents pre-loaded with your work context so recurring sessions resume instantly instead of cold-starting. | Use BlitzOS to choose an implementation surface for repeatable agent work. | Inspectable GitHub source (142 stars; 17 forks; MIT license; updated 2026-07-27); popularity is context, not proof of reliability. | medium | README.md | 928 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L928 | 2026-07-22 | Build | build | Choose runtimes, tools, and delegation surfaces. | context | builder | agent | direct | source-implementation | A | ok | https://github.com/blitzdotdev/blitzos | GitHub - blitzdotdev/blitzos: BlitzOS allows cloud agents to boot already knowing your work, and keep working with your laptop closed · GitHub | BlitzOS allows cloud agents to boot already knowing your work, and keep working with your laptop closed - blitzdotdev/blitzos | 2026-07-13 | 2026 | blitzdotdev/blitzos | GitHub | github-api | blitzdotdev/blitzos | 142 | 17 | MIT | 2026-07-13T23:44:04Z | 2026-07-27T07:27:44Z | 2026-07-29T08:06:54 | |||||
ale-0304 | Coding-Agent Loop Systems | coding-agent-loop-systems | Blog | 📝 | Star Fleet Math: Solving Erdős Problems with 20 Parallel Codex Harnesses | https://www.starfleetmath.com/ | external | www.starfleetmath.com | Field report on running twenty parallel Codex harnesses against open Erdős problems, a live demonstration of fleet-scale loop orchestration applied to mathematics research. | Field report on running twenty parallel Codex harnesses against open Erdős problems, a live demonstration of fleet-scale loop orchestration applied to mathematics research. | Field report on running twenty parallel Codex harnesses against open Erdős problems, a live demonstration of fleet-scale loop orchestration applied to mathematics research. | Orchestration and control flow are made explicit and inspectable. Field report on running twenty parallel Codex harnesses against open Erdős problems, a live demonstration of fleet-scale loop orchestration applied to mathematics research. | Use Star Fleet Math: Solving Erdős Problems with 20 Parallel Codex Harnesses to choose an implementation surface for repeatable agent work. | Contextual source from www.starfleetmath.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 929 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L929 | 2026-07-22 | Build | build | Choose runtimes, tools, and delegation surfaces. | delegation | builder | agent | direct | practitioner-analysis | B | ok | https://www.starfleetmath.com/ | Star Fleet Math | A Mac app controlling up to 20 starship agentic harnesses running GPT-5.6 Sol max in parallel on open mathematics problems. | Star Fleet Math | html-meta | 2026-07-29T08:06:54 | ||||||||||||||
ale-0305 | Coding-Agent Loop Systems | coding-agent-loop-systems | Tool | 🧰 | Finn-loop | https://github.com/finna/Finn-loop | external | github.com | Three Claude Code skills from Alex Finn that turn Linear + GitHub into a small, human-gated software factory: /finn-spec interviews you until the behavior is unambiguous and files a Linear issue with acceptance criteria and non-goals, /finn-build claims the next agent-ready issue and opens a PR (run repeatedly via the ... | Three Claude Code skills from Alex Finn that turn Linear + GitHub into a small, human-gated software factory: /finn-spec interviews you until the behavior is unambiguous and files a Linear issue with acceptance criteria and non-goals, /finn-build claims the next agent-ready issue and opens a PR (run repeatedly via the ... | Three Claude Code skills from Alex Finn that turn Linear + GitHub into a small, human-gated software factory: /finn-spec interviews you until the behavior is unambiguous and files a Linear issue with acceptance criteria and non-goals, /finn-build claims the next agent-ready issue and opens a PR (run repeatedly via the ... | Uses real automated software-engineering systems as evidence for practical loop architectures. Three Claude Code skills from Alex Finn that turn Linear + GitHub into a small, human-gated software factory: /finn-spec interviews you until the behavior is unambiguous and files a Linear issue with acceptance criteria and n... | Use Finn-loop to choose an implementation surface for repeatable agent work. | Inspectable GitHub source (233 stars; 39 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 930 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L930 | 2026-07-25 | Build | build | Choose runtimes, tools, and delegation surfaces. | objective;intake;escalation | builder | agent | direct | source-implementation | A | ok | https://github.com/finna/Finn-loop | GitHub - finna/Finn-loop: The Finn-loop: a 3-skill AI software factory for Claude Code — spec, build, review. Humans merge. · GitHub | The Finn-loop: a 3-skill AI software factory for Claude Code — spec, build, review. Humans merge. - finna/Finn-loop | 2026-07-22 | 2026 | finna/Finn-loop | GitHub | github-api | finna/Finn-loop | 233 | 39 | MIT | 2026-07-22T16:48:50Z | 2026-07-29T06:29:15Z | 2026-07-29T08:06:54 | |||||
ale-0306 | Coding-Agent Loop Systems | coding-agent-loop-systems | Paper | 📄 | How Do AI Coding Agents Contribute to Software Development? An Empirical Study of Agentic Pull Requests | https://arxiv.org/abs/2607.21832 | external | arxiv.org | IN WINDOW (announced Jul 28, 2026; submitted Jul 23). Mazloomzadeh, Morovati, and Khomh run a longitudinal analysis over the AIDev dataset covering 9,428 agentic PRs from five agents across 489 Python repositories. Measured, partly counter-narrative results: merge rates vary sharply by agent (Claude 84.3%, Codex 73.5%,... | IN WINDOW (announced Jul 28, 2026; submitted Jul 23). Mazloomzadeh, Morovati, and Khomh run a longitudinal analysis over the AIDev dataset covering 9,428 agentic PRs from five agents across 489 Python repositories. Measured, partly counter-narrative results: merge rates vary sharply by agent (Claude 84.3%, Codex 73.5%,... | IN WINDOW (announced Jul 28, 2026; submitted Jul 23). Mazloomzadeh, Morovati, and Khomh run a longitudinal analysis over the AIDev dataset covering 9,428 agentic PRs from five agents across 489 Python repositories. Measured, partly counter-narrative results: merge rates vary sharply by agent (Claude 84.3%, Codex 73.5%,... | Packages the evidence as queryable CSV and JSONL rather than only a rendered page. IN WINDOW (announced Jul 28, 2026; submitted Jul 23). Mazloomzadeh, Morovati, and Khomh run a longitudinal analysis over the AIDev dataset covering 9,428 agentic PRs from five agents across 489 Python repositories. Measured, partly count... | Use How Do AI Coding Agents Contribute to Software Development? An Empirical Study of Agentic Pull Requests to choose an implementation surface for repeatable agent work. | Research source arXiv:2607.21832; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 931 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L931 | 2026-07-28 | Build | build | Choose runtimes, tools, and delegation surfaces. | verification;escalation | researcher;evaluator | agent | direct | research-preprint | A | ok | https://arxiv.org/abs/2607.21832 | [2607.21832] How Do AI Coding Agents Contribute to Software Development? an Empirical Study of Agentic Pull Requests | Recent advances in large language models and their rapid adoption across software engineering tasks have made Artificial Intelligence (AI) coding agents an integral component of modern software development workflows. While developers increasingly benefit from these coding agents, their impact on software quality remain... | Iren Mazloomzadeh; Mohammad Mehdi Morovati; Foutse Khomh | 2026-07-23 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.21832 | 2026-07-29T08:06:54 | ||||||||
ale-0307 | Coding-Agent Loop Systems | coding-agent-loop-systems | Paper | 📄 | Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Coding Agent Teams | https://arxiv.org/abs/2607.22917 | external | arxiv.org | Explicitly diagnoses four Claude Code Agent Teams failure modes and builds a persistent workspace to fix them: irrecoverable agent teams (state lost when the terminal closes), compaction eroding working detail, agentic 'technical debt' trapped in compacted old chats, and heavy handoff prompt writing. This is loop engin... | Explicitly diagnoses four Claude Code Agent Teams failure modes and builds a persistent workspace to fix them: irrecoverable agent teams (state lost when the terminal closes), compaction eroding working detail, agentic 'technical debt' trapped in compacted old chats, and heavy handoff prompt writing. This is loop engin... | Explicitly diagnoses four Claude Code Agent Teams failure modes and builds a persistent workspace to fix them: irrecoverable agent teams (state lost when the terminal closes), compaction eroding working detail, agentic 'technical debt' trapped in compacted old chats, and heavy handoff prompt writing. This is loop engin... | Durable execution and replay are treated as first-class loop infrastructure. Explicitly diagnoses four Claude Code Agent Teams failure modes and builds a persistent workspace to fix them: irrecoverable agent teams (state lost when the terminal closes), compaction eroding working detail, agentic 'technical debt' trapped... | Use Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Coding Agent Teams to choose an implementation surface for repeatable agent work. | Research source arXiv:2607.22917; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 932 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L932 | 2026-07-28 | Build | build | Choose runtimes, tools, and delegation surfaces. | workspace;delegation;state;escalation | researcher;evaluator | agent | direct | research-preprint | A | ok | https://arxiv.org/abs/2607.22917 | [2607.22917] Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Coding Agent Teams | Large Language Model (LLM) agents have significantly improved coding and programming workflows. Claude Code, in particular, is one of the most powerful LLM coding agents and is capable of conducting complex coding tasks. However, several drawbacks can undermine long-term agentic workflows. (1) Irrecoverable agent teams... | Shouren Wang | 2026-07-24 | 2026 | arXiv | arXiv | 31 pages, 9 figures | cs.AI | arxiv-api | 2607.22917 | 2026-07-29T08:06:54 | |||||||
ale-0308 | Coding-Agent Loop Systems | coding-agent-loop-systems | Paper | 📄 | Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents | https://arxiv.org/abs/2607.21909 | external | arxiv.org | Reframes parallel coding agents as a pre-write admission problem instead of a merge-time repair problem. Each worker declares a versioned ChangeIntent (exact base commit, typed resources, dependencies, operations marked committed or contingent) and a deterministic control plane atomically admits compatible intents, con... | Reframes parallel coding agents as a pre-write admission problem instead of a merge-time repair problem. Each worker declares a versioned ChangeIntent (exact base commit, typed resources, dependencies, operations marked committed or contingent) and a deterministic control plane atomically admits compatible intents, con... | Reframes parallel coding agents as a pre-write admission problem instead of a merge-time repair problem. Each worker declares a versioned ChangeIntent (exact base commit, typed resources, dependencies, operations marked committed or contingent) and a deterministic control plane atomically admits compatible intents, con... | Uses real automated software-engineering systems as evidence for practical loop architectures. Reframes parallel coding agents as a pre-write admission problem instead of a merge-time repair problem. Each worker declares a versioned ChangeIntent (exact base commit, typed resources, dependencies, operations marked commi... | Use Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents to choose an implementation surface for repeatable agent work. | Research source arXiv:2607.21909; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 933 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L933 | 2026-07-28 | Build | build | Choose runtimes, tools, and delegation surfaces. | workspace;delegation;verification;state | researcher;evaluator | agent | direct | research-preprint | A | ok | https://arxiv.org/abs/2607.21909 | [2607.21909] Claim Plane: Enforceable Change Intents and Dynamic Scope for Parallel Coding Agents | Parallel coding agents can independently produce locally valid changes while still interfering at integration time, expanding beyond planned scope, or relying on premises invalidated by concurrent work. Existing responses emphasize communication, isolated workspaces, late merge-time repair, continuous supervision, or p... | Maxim Nikolaev | 2026-07-24 | 2026 | arXiv | arXiv | 10 pages, 2 figures. Preprint | cs.SE | arxiv-api | 2607.21909 | 2026-07-29T08:06:54 | |||||||
ale-0309 | Coding-Agent Loop Systems | coding-agent-loop-systems | Paper | 📄 | Plans Work in Mysterious Ways: Evaluating a Plan Mode for Spreadsheet Agents | https://arxiv.org/abs/2607.23670 | external | arxiv.org | Plan Mode is now standard in agentic coding tools, but nobody has tested whether the transparency-and-control benefit survives outside developer contexts. A within-subjects study (N=24) with a spreadsheet Plan Mode prototype against a non-planning baseline finds task outcomes essentially unchanged, yet Plan Mode reduce... | Plan Mode is now standard in agentic coding tools, but nobody has tested whether the transparency-and-control benefit survives outside developer contexts. A within-subjects study (N=24) with a spreadsheet Plan Mode prototype against a non-planning baseline finds task outcomes essentially unchanged, yet Plan Mode reduce... | Plan Mode is now standard in agentic coding tools, but nobody has tested whether the transparency-and-control benefit survives outside developer contexts. A within-subjects study (N=24) with a spreadsheet Plan Mode prototype against a non-planning baseline finds task outcomes essentially unchanged, yet Plan Mode reduce... | Uses real automated software-engineering systems as evidence for practical loop architectures. Plan Mode is now standard in agentic coding tools, but nobody has tested whether the transparency-and-control benefit survives outside developer contexts. A within-subjects study (N=24) with a spreadsheet Plan Mode prototype ... | Use Plans Work in Mysterious Ways: Evaluating a Plan Mode for Spreadsheet Agents to choose an implementation surface for repeatable agent work. | Research source arXiv:2607.23670; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 934 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L934 | 2026-07-28 | Build | build | Choose runtimes, tools, and delegation surfaces. | workspace;verification;escalation | researcher;evaluator | agent | direct | research-preprint | A | ok | https://arxiv.org/abs/2607.23670 | [2607.23670] Plans Work in Mysterious Ways: Evaluating a Plan Mode for Spreadsheet Agents | Plan Modes have become standard features in agentic programming tools, allowing users to gain transparency and control by working with the agent to develop a plan before task execution. However, it remains unclear whether the benefits of this feature translate to end-user programming environments such as spreadsheets. ... | Aayush Kumar; Avik Dutta; Sumit Gulwani; Gustavo Soares; Advait Sarkar; Emerson Murphy-Hill | 2026-07-26 | 2026 | arXiv | arXiv | cs.HC | arxiv-api | 2607.23670 | 2026-07-29T08:06:54 | ||||||||
ale-0310 | Coding-Agent Loop Systems | coding-agent-loop-systems | Tool | 🧰 | JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents | https://arxiv.org/abs/2607.23588 | external | arxiv.org | Open harness for long-horizon multimodal production, built on the observation that real creative work is an evolving project state -- references, drafts, alternatives, edits, failed attempts, version relations, tool actions, evaluation signals, human feedback -- that prompt-based, chat-based, and node-based systems dis... | Open harness for long-horizon multimodal production, built on the observation that real creative work is an evolving project state -- references, drafts, alternatives, edits, failed attempts, version relations, tool actions, evaluation signals, human feedback -- that prompt-based, chat-based, and node-based systems dis... | Open harness for long-horizon multimodal production, built on the observation that real creative work is an evolving project state -- references, drafts, alternatives, edits, failed attempts, version relations, tool actions, evaluation signals, human feedback -- that prompt-based, chat-based, and node-based systems dis... | Evaluation data is used as the feedback signal for improving loop behavior. Open harness for long-horizon multimodal production, built on the observation that real creative work is an evolving project state -- references, drafts, alternatives, edits, failed attempts, version relations, tool actions, evaluation signals,... | Use JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents to choose an implementation surface for repeatable agent work. | Research source arXiv:2607.23588; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 935 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L935 | 2026-07-28 | Build | build | Choose runtimes, tools, and delegation surfaces. | workspace;verification;state;escalation | builder | agent | direct | research-preprint | A | ok | https://arxiv.org/abs/2607.23588 | [2607.23588] JarvisHub: An Open Harness for Canvas-Native Multimodal Creative Agents | Creative AI is moving from single-step asset generation toward long-horizon multimodal production. Although recent generative models can synthesize high-quality images, videos, audio clips, UI elements, storyboards, slides, and other creative assets, real-world creative work requires more than isolated prompt-output in... | Yunlong Lin; Zixu Lin; Zhaohu Xing; Biqiang Li; Chenxin Li; Haonan Wang; Haitao Wu; Hengyu Liu; Jianghai Chen; Kaituo Feng; Kaixin Li; Shawn Chen; Shijue Huang; Sixiang Chen; Tsung-Yi Ho; Wenxuan Huang; Xiangyan Liu; Xiaomeng Hu; Xuanhua He; Yan Sun; Yunqing Zhao; Zhiqin Yang; Zehan Wang; Zhengyang Tang; Tianyu Pang; X... | 2026-07-26 | 2026 | arXiv | arXiv | 15 pages, 9 figures. Project page: https://www.jarvishub.site/ Code github: https://github.com/LYL1015/JarvisHub | cs.CV | arxiv-api | 2607.23588 | 2026-07-29T08:06:54 | |||||||
ale-0311 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Why Agentic Systems Must Produce Deterministic Outputs to Scale | https://streamzero.com/blog/posts/deep-dives-tools-technologies-architectures/agentic-patterns/why-agentic-systems-must-produce-deterministic-outputs-to-scale | external | streamzero.com | Argues for deterministic boundaries, contracts, and execution gates around probabilistic agent reasoning. | Argues for deterministic boundaries, contracts, and execution gates around probabilistic agent reasoning. | Argues for deterministic boundaries, contracts, and execution gates around probabilistic agent reasoning. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Argues for deterministic boundaries, contracts, and execution gates around probabilistic agent reasoning. | Use Why Agentic Systems Must Produce Deterministic Outputs to Scale to measure progress and gate completion with repeatable evidence. | Contextual source from streamzero.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 944 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L944 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://streamzero.com/blog/posts/deep-dives-tools-technologies-architectures/agentic-patterns/why-agentic-systems-must-produce-deterministic-outputs-to-scale | Why Agentic Systems Must Produce Deterministic Outputs to Scale | Agentic systems are gaining traction, but their inherent non-determinism poses a significant challenge for production environments. This document argues that deterministic outputs are essential for scaling agentic systems, enabling validation, security, and compliance in critical applications. | streamzero.com | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0312 | Verification And Feedback Gates | verification-and-feedback-gates | Pattern | 🔁 | Stop Babysitting Your Coding Agent. Give It Backpressure. | https://generativeprogrammer.com/p/stop-babysitting-your-coding-agent | external | generativeprogrammer.com | Explains how to turn tests, linters, builds, traces, and other signals into feedback loops for coding agents. | Explains how to turn tests, linters, builds, traces, and other signals into feedback loops for coding agents. | Explains how to turn tests, linters, builds, traces, and other signals into feedback loops for coding agents. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Explains how to turn tests, linters, builds, traces, and other signals into feedback loops for coding agents. | Use Stop Babysitting Your Coding Agent. Give It Backpressure. to measure progress and gate completion with repeatable evidence. | Operational pattern or playbook; signal comes from reusable loop structure and practical transferability. | medium | README.md | 945 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L945 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;exit | builder;evaluator | harness | enabling | operational-pattern | B | ok | https://generativeprogrammer.com/p/stop-babysitting-your-coding-agent | Stop Babysitting Your Coding Agent. Give It Backpressure. | Backpressure is feedback that reaches the agent before the agent reaches the human. | Bilgin Ibryam | generativeprogrammer.com | html-meta | 2026-07-29T08:06:54 | ||||||||||||||
ale-0313 | Verification And Feedback Gates | verification-and-feedback-gates | Pattern | 🔁 | How to Build a Self-Verification Loop in Claude Code | https://dev.to/shipwithaiio/how-to-build-a-self-verification-loop-in-claude-code-3-layers-20-minutes-m1p | external | dev.to | Uses hooks to enforce syntax, intent, and regression checks before an agent can finish. | Uses hooks to enforce syntax, intent, and regression checks before an agent can finish. | Uses hooks to enforce syntax, intent, and regression checks before an agent can finish. | The agent workflow includes explicit self-checking or gated completion. Uses hooks to enforce syntax, intent, and regression checks before an agent can finish. | Use How to Build a Self-Verification Loop in Claude Code to measure progress and gate completion with repeatable evidence. | Operational pattern or playbook; signal comes from reusable loop structure and practical transferability. | medium | README.md | 946 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L946 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | operational-pattern | B | ok | https://dev.to/shipwithaiio/how-to-build-a-self-verification-loop-in-claude-code-3-layers-20-minutes-m1p | How to Build a Self-Verification Loop in Claude Code (3 Layers, 20 Minutes) - DEV Community Navigation menu Search Search Close More... Copy link Enter fullscreen mode Exit fullscreen mode Enter fullscreen mode Exit fullscreen mode Enter fullscreen mode Exit fullscreen mode Enter fullscreen mode Exit fullscreen mode En... | Claude Code's Stop hook blocks the agent from finishing until verification passes. Combine it with... Tagged with ai, programming, productivity, claude. | DEV Community | html-meta | 2026-07-29T08:06:54 | |||||||||||||||
ale-0314 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Agentic Code Review | https://addyosmani.com/blog/agentic-code-review/ | external | addyosmani.com | Addy Osmani argues that review, not code generation, is the bottleneck in agentic workflows, proposing risk-tiered verification depth, heterogeneous AI reviewers, and hard CI gates while warning against closed loops of models with correlated blind spots. | Addy Osmani argues that review, not code generation, is the bottleneck in agentic workflows, proposing risk-tiered verification depth, heterogeneous AI reviewers, and hard CI gates while warning against closed loops of models with correlated blind spots. | Addy Osmani argues that review, not code generation, is the bottleneck in agentic workflows, proposing risk-tiered verification depth, heterogeneous AI reviewers, and hard CI gates while warning against closed loops of models with correlated blind spots. | Verification is promoted from a final check to a loop-control signal. Addy Osmani argues that review, not code generation, is the bottleneck in agentic workflows, proposing risk-tiered verification depth, heterogeneous AI reviewers, and hard CI gates while warning against closed loops of models with correlated blind sp... | Use Agentic Code Review to measure progress and gate completion with repeatable evidence. | Contextual source from addyosmani.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 947 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L947 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://addyosmani.com/blog/agentic-code-review/ | AddyOsmani.com - Agentic Code Review | Coding agents are extraordinarily good now, and getting better fast. The interesting consequence is that the hard part of engineering moved from writing code... | Addy Osmani | addyosmani.com | html-meta | 2026-07-29T08:06:54 | ||||||||||||||
ale-0315 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Using DSPy to Evaluate and Improve Datasette Agent's SQL System Prompts | https://simonwillison.net/2026/Jul/2/dspy-datasette-agent-prompts/ | external | simonwillison.net | Simon Willison wires a DSPy evaluation harness to a live Datasette instance with real tool calls and gold-standard metrics, then uses the eval traces to find and fix weaknesses in the agent's SQL system prompt. | Simon Willison wires a DSPy evaluation harness to a live Datasette instance with real tool calls and gold-standard metrics, then uses the eval traces to find and fix weaknesses in the agent's SQL system prompt. | Simon Willison wires a DSPy evaluation harness to a live Datasette instance with real tool calls and gold-standard metrics, then uses the eval traces to find and fix weaknesses in the agent's SQL system prompt. | Evaluation data is used as the feedback signal for improving loop behavior. Simon Willison wires a DSPy evaluation harness to a live Datasette instance with real tool calls and gold-standard metrics, then uses the eval traces to find and fix weaknesses in the agent's SQL system prompt. | Use Using DSPy to Evaluate and Improve Datasette Agent's SQL System Prompts to measure progress and gate completion with repeatable evidence. | Contextual source from simonwillison.net; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 948 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L948 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://simonwillison.net/2026/Jul/2/dspy-datasette-agent-prompts/ | Research: Using DSPy to evaluate and improve Datasette Agent's SQL system prompts | Leveraging the DSPy framework, this project evaluates and refines the core production system prompts used by Datasette Agent’s read-only SQL question answerer. The methodology involves a harness where DSPy agents … | Simon Willison | 2026 | Simon Willison’s Weblog | html-meta | 2026-07-29T08:06:54 | |||||||||||||
ale-0316 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Agentic coding notes | https://danluu.com/ai-coding/ | external | danluu.com | Dan Luu's first-hand benchmarks and workflows arguing that systematic test infrastructure such as fuzzing and randomized testing, not human review, is what lets agent-generated code ship, and documenting why a self-contained agentic quality loop has so far eluded him. | Dan Luu's first-hand benchmarks and workflows arguing that systematic test infrastructure such as fuzzing and randomized testing, not human review, is what lets agent-generated code ship, and documenting why a self-contained agentic quality loop has so far eluded him. | Dan Luu's first-hand benchmarks and workflows arguing that systematic test infrastructure such as fuzzing and randomized testing, not human review, is what lets agent-generated code ship, and documenting why a self-contained agentic quality loop has so far eluded him. | The work turns loop quality into a measurable task or score. Dan Luu's first-hand benchmarks and workflows arguing that systematic test infrastructure such as fuzzing and randomized testing, not human review, is what lets agent-generated code ship, and documenting why a self-contained agentic quality loop has so far el... | Use Agentic coding notes to measure progress and gate completion with repeatable evidence. | Contextual source from danluu.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 949 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L949 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | evaluator | harness | enabling | practitioner-analysis | B | ok | https://danluu.com/ai-coding/ | Agentic test processes, LLM benchmarks, and other notes on agentic coding from Galapagos Island | danluu.com | domain-fallback | 2026-07-29T08:06:54 | ||||||||||||||||
ale-0317 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Understanding Is the New Bottleneck | https://www.geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html | external | www.geoffreylitt.com | Geoffrey Litt argues that human understanding, not verification, is the real bottleneck in agent loops, warning that cognitive debt accrues when iterations outpace comprehension and proposing literate diffs, quizzes, and interactive micro-worlds as speed regulators. | Geoffrey Litt argues that human understanding, not verification, is the real bottleneck in agent loops, warning that cognitive debt accrues when iterations outpace comprehension and proposing literate diffs, quizzes, and interactive micro-worlds as speed regulators. | Geoffrey Litt argues that human understanding, not verification, is the real bottleneck in agent loops, warning that cognitive debt accrues when iterations outpace comprehension and proposing literate diffs, quizzes, and interactive micro-worlds as speed regulators. | Verification is promoted from a final check to a loop-control signal. Geoffrey Litt argues that human understanding, not verification, is the real bottleneck in agent loops, warning that cognitive debt accrues when iterations outpace comprehension and proposing literate diffs, quizzes, and interactive micro-worlds as s... | Use Understanding Is the New Bottleneck to measure progress and gate completion with repeatable evidence. | Contextual source from www.geoffreylitt.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 950 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L950 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | evaluator | harness | enabling | practitioner-analysis | B | ok | https://www.geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html | Understanding is the new bottleneck | Agents can write code faster than we can absorb it. Here's why it still matters for humans to understand what they build — and some techniques for doing that efficiently: explainer docs, quizzes, micro-worlds, and shared spaces. | 2026 | geoffreylitt.com | url-date | 2026-07-29T08:06:54 | ||||||||||||||
ale-0318 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Verifying Agentic Development at Scale | https://cognition.com/blog/testing-development | external | cognition.com | Cognition details the verification stack behind Devin sessions going majority-async: source-grounded test plans, deterministic reusable testing skills, and annotated video artifacts with pass/fail assertions so unattended runs return merge-ready results. | Cognition details the verification stack behind Devin sessions going majority-async: source-grounded test plans, deterministic reusable testing skills, and annotated video artifacts with pass/fail assertions so unattended runs return merge-ready results. | Cognition details the verification stack behind Devin sessions going majority-async: source-grounded test plans, deterministic reusable testing skills, and annotated video artifacts with pass/fail assertions so unattended runs return merge-ready results. | Verification is promoted from a final check to a loop-control signal. Cognition details the verification stack behind Devin sessions going majority-async: source-grounded test plans, deterministic reusable testing skills, and annotated video artifacts with pass/fail assertions so unattended runs return merge-ready resu... | Use Verifying Agentic Development at Scale to measure progress and gate completion with repeatable evidence. | Contextual source from cognition.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 951 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L951 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://cognition.com/blog/testing-development | Verifying Agentic Development at Scale | Cognition | What we’ve learned building end-to-end testing capabilities in Devin’s virtual machine | 2026-05-29 | 2026 | cognition.com | html-meta | 2026-07-29T08:06:54 | |||||||||||||
ale-0319 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Loop Engineering Without Verification Is Just Automation | https://www.sonarsource.com/blog/loop-engineering-without-verification-is-just-automation/ | external | www.sonarsource.com | Sonar formalizes a two-tier verification gate for agent loops, pairing a probabilistic LLM verifier sub-agent for intent with a deterministic analysis gate as the hard halt, arguing that LLM-only verification amounts to two optimists agreeing. | Sonar formalizes a two-tier verification gate for agent loops, pairing a probabilistic LLM verifier sub-agent for intent with a deterministic analysis gate as the hard halt, arguing that LLM-only verification amounts to two optimists agreeing. | Sonar formalizes a two-tier verification gate for agent loops, pairing a probabilistic LLM verifier sub-agent for intent with a deterministic analysis gate as the hard halt, arguing that LLM-only verification amounts to two optimists agreeing. | Verification is promoted from a final check to a loop-control signal. Sonar formalizes a two-tier verification gate for agent loops, pairing a probabilistic LLM verifier sub-agent for intent with a deterministic analysis gate as the hard halt, arguing that LLM-only verification amounts to two optimists agreeing. | Use Loop Engineering Without Verification Is Just Automation to measure progress and gate completion with repeatable evidence. | Contextual source from www.sonarsource.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 952 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L952 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://www.sonarsource.com/blog/loop-engineering-without-verification-is-just-automation/ | Loop engineering without verification is just automation | Sonar | Explore how LLM reviewers and deterministic checks work together to keep coding agent loops from shipping unfinished code. | sonarsource.com | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0320 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Closing the Verification Loop: Observability-Driven Harnesses | https://www.datadoghq.com/blog/ai/harness-first-agents/ | external | www.datadoghq.com | Datadog engineers' case for harness-first engineering once agents write code faster than humans can review, using deterministic simulation testing across millions of seeds as the verification gate. | Datadog engineers' case for harness-first engineering once agents write code faster than humans can review, using deterministic simulation testing across millions of seeds as the verification gate. | Datadog engineers' case for harness-first engineering once agents write code faster than humans can review, using deterministic simulation testing across millions of seeds as the verification gate. | Verification is promoted from a final check to a loop-control signal. Datadog engineers' case for harness-first engineering once agents write code faster than humans can review, using deterministic simulation testing across millions of seeds as the verification gate. | Use Closing the Verification Loop: Observability-Driven Harnesses to measure progress and gate completion with repeatable evidence. | Contextual source from www.datadoghq.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 953 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L953 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://www.datadoghq.com/blog/ai/harness-first-agents/ | Closing the verification loop: Observability-driven harnesses for building with agents | Datadog security-platform rum ci dashboard host-map apm security-platform rum ci dashboard host-map apm security-platform rum ci dashboard Icon/world | Learn how Datadog verifies AI-generated systems at scale using deterministic testing, formal methods, and observability-driven feedback loops. | Alp Keles, Jai Menon, Sesh Nalla, Vyom Shah | 2026-03-09 | 2026 | Datadog | html-meta | 2026-07-29T08:06:54 | ||||||||||||
ale-0321 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | How to build a better agent harness with traces and evals | https://arize.com/blog/improve-ai-agents-traces-evals-harness/ | external | arize.com | Trace-evaluate-debug-refine loop for improving agent behavior from real runs. | Trace-evaluate-debug-refine loop for improving agent behavior from real runs. | Trace-evaluate-debug-refine loop for improving agent behavior from real runs. | Evaluation data is used as the feedback signal for improving loop behavior. Trace-evaluate-debug-refine loop for improving agent behavior from real runs. | Use How to build a better agent harness with traces and evals to measure progress and gate completion with repeatable evidence. | Contextual source from arize.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 958 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L958 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://arize.com/blog/improve-ai-agents-traces-evals-harness/ | How to build a better agent harness with traces and evals - Arize AI | Agents are easy to prototype and hard to improve. A repeatable loop of traces, evals, failed-span inspection, and targeted harness changes makes agent behavior easier to debug and improve. | Aaron Winston | 2026-05-29 | 2026 | Arize AI | html-meta | 2026-07-29T08:06:54 | ||||||||||||
ale-0322 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Better Harness: A Recipe for Harness Hill-Climbing with Evals | https://www.langchain.com/blog/better-harness-a-recipe-for-harness-hill-climbing-with-evals | external | www.langchain.com | LangChain's recipe for using evals as the learning signal for harness improvement. | LangChain's recipe for using evals as the learning signal for harness improvement. | LangChain's recipe for using evals as the learning signal for harness improvement. | Evaluation data is used as the feedback signal for improving loop behavior. LangChain's recipe for using evals as the learning signal for harness improvement. | Use Better Harness: A Recipe for Harness Hill-Climbing with Evals to measure progress and gate completion with repeatable evidence. | Contextual source from www.langchain.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 959 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L959 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://www.langchain.com/blog/better-harness-a-recipe-for-harness-hill-climbing-with-evals | Better Harness: A Recipe for Harness Hill-Climbing with Evals | We can build better agents by building better harnesses. But to autonomously build a “better” harness, we need a strong learning signal to “hill-climb” on. We share how we use evals as that signal, plus design decisions that help our agent generalize instead of overfit. Better-Harness is a system for iteratively sourci... | LangChain | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0323 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | Improving Deep Agents with harness engineering | https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering | external | www.langchain.com | Practical discussion of self-verification, traces, middleware, and loop detection for coding agents. | Practical discussion of self-verification, traces, middleware, and loop detection for coding agents. | Practical discussion of self-verification, traces, middleware, and loop detection for coding agents. | The agent workflow includes explicit self-checking or gated completion. Practical discussion of self-verification, traces, middleware, and loop detection for coding agents. | Use Improving Deep Agents with harness engineering to measure progress and gate completion with repeatable evidence. | Contextual source from www.langchain.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 960 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L960 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering | Improving Deep Agents with harness engineering | Harness engineering improved LangChain's coding agent from Top 30 to Top 5 on Terminal Bench using self-verification, tracing, and context optimization. | LangChain | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0324 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses | https://arxiv.org/abs/2604.25850 | external | arxiv.org | Closed loop that turns each harness edit into a falsifiable contract verified against trajectory outcomes, so the harness evolves from observability rather than trial and error. | Closed loop that turns each harness edit into a falsifiable contract verified against trajectory outcomes, so the harness evolves from observability rather than trial and error. | Closed loop that turns each harness edit into a falsifiable contract verified against trajectory outcomes, so the harness evolves from observability rather than trial and error. | Verification is promoted from a final check to a loop-control signal. Closed loop that turns each harness edit into a falsifiable contract verified against trajectory outcomes, so the harness evolves from observability rather than trial and error. | Use Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses to measure progress and gate completion with repeatable evidence. | Research source arXiv:2604.25850; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 961 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L961 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2604.25850 | [2604.25850] Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses | Harnesses are now central to coding-agent performance, mediating how models interact with tools and execution environments. Yet harness engineering remains a manual craft, because automating it faces a heterogeneous action space across editable components, voluminous trajectories that bury actionable signal, and edits ... | Jiahang Lin; Shichun Liu; Chengjun Pan; Lizhi Lin; Shihan Dou; Zhiheng Xi; Xuanjing Huang; Hang Yan; Zhenhua Han; Tao Gui; Yu-Gang Jiang | 2026-04-28 | 2026 | arXiv | arXiv | cs.CL | arxiv-api | 2604.25850 | 2026-07-29T08:06:54 | |||||||||
ale-0325 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Meta-Harness: End-to-End Optimization of Model Harnesses | https://arxiv.org/abs/2603.28052 | external | arxiv.org | Optimizes the surrounding harness (tools, prompts, control flow) end to end against task outcomes, turning harness tuning into a measurable improvement loop instead of manual trial and error. | Optimizes the surrounding harness (tools, prompts, control flow) end to end against task outcomes, turning harness tuning into a measurable improvement loop instead of manual trial and error. | Optimizes the surrounding harness (tools, prompts, control flow) end to end against task outcomes, turning harness tuning into a measurable improvement loop instead of manual trial and error. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Optimizes the surrounding harness (tools, prompts, control flow) end to end against task outcomes, turning harness tuning into a measurable improvement loop instead of manual trial and error. | Use Meta-Harness: End-to-End Optimization of Model Harnesses to measure progress and gate completion with repeatable evidence. | Research source arXiv:2603.28052; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 962 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L962 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2603.28052 | [2603.28052] Meta-Harness: End-to-End Optimization of Model Harnesses | The performance of large language model (LLM) systems depends not only on model weights, but also on their harness: the code that determines what information to store, retrieve, and present to the model. Yet harnesses are still designed largely by hand, and existing text optimizers are poorly matched to this setting be... | Yoonho Lee; Roshen Nair; Qizheng Zhang; Kangwook Lee; Omar Khattab; Chelsea Finn | 2026-03-30 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2603.28052 | 2026-07-29T08:06:54 | |||||||||
ale-0326 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | HALO (Hierarchical Agent Loop Optimizer) | https://github.com/context-labs/halo | external | github.com | Analyzes production agent traces to find harness-level failure modes, hands its report to a coding agent to apply fixes, and repeats the collect-analyze-fix-redeploy cycle, reporting AppWorld gains from harness changes alone. | Analyzes production agent traces to find harness-level failure modes, hands its report to a coding agent to apply fixes, and repeats the collect-analyze-fix-redeploy cycle, reporting AppWorld gains from harness changes alone. | Analyzes production agent traces to find harness-level failure modes, hands its report to a coding agent to apply fixes, and repeats the collect-analyze-fix-redeploy cycle, reporting AppWorld gains from harness changes alone. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Analyzes production agent traces to find harness-level failure modes, hands its report to a coding agent to apply fixes, and repeats the collect-analyze-fix-redeploy cycle, reporting AppWorld gains from harness changes alone. | Use HALO (Hierarchical Agent Loop Optimizer) to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (1,135 stars; 84 forks; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 963 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L963 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/context-labs/halo | GitHub - context-labs/HALO: Hierarchal Agent Loop Optimizer · GitHub | Hierarchal Agent Loop Optimizer. Contribute to context-labs/HALO development by creating an account on GitHub. | 2026-04-21 | 2026 | context-labs/halo | GitHub | github-api | context-labs/halo | 1135 | 84 | 2026-04-21T18:20:46Z | 2026-07-29T01:34:13Z | 2026-07-29T08:06:54 | |||||||
ale-0327 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Harness-Aware Self-Evolving: Co-Evolving Model Weights, Harness, and Task Solutions | https://arxiv.org/abs/2607.03935 | external | arxiv.org | Agentic RL framework in which one model both solves tasks and edits its own harness, including repairing faulty evaluation code, co-evolving weights, harness, and solutions so a trained Qwen3-8B matches a much larger baseline. | Agentic RL framework in which one model both solves tasks and edits its own harness, including repairing faulty evaluation code, co-evolving weights, harness, and solutions so a trained Qwen3-8B matches a much larger baseline. | Agentic RL framework in which one model both solves tasks and edits its own harness, including repairing faulty evaluation code, co-evolving weights, harness, and solutions so a trained Qwen3-8B matches a much larger baseline. | Evaluation data is used as the feedback signal for improving loop behavior. Agentic RL framework in which one model both solves tasks and edits its own harness, including repairing faulty evaluation code, co-evolving weights, harness, and solutions so a trained Qwen3-8B matches a much larger baseline. | Use Harness-Aware Self-Evolving: Co-Evolving Model Weights, Harness, and Task Solutions to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.03935; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 964 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L964 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.03935 | [2607.03935] Harness-Aware Self-Evolving: Co-Evolving Model Weights, Harness, and Task Solutions | Self-evolving frameworks usually optimize task solutions while treating the surrounding harness as fixed. We introduce Harness-Aware Self-Evolving (HASE), an agentic reinforcement-learning framework in which a single model can generate task solutions or edit selected harness components in a multi-turn action space. HAS... | Haochen Luo; Yi Huang; Sichun Luo; Fengyuan Liu; Lei Li; Zefa Hu; Junlan Feng; Qi Liu | 2026-07-04 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.03935 | 2026-07-29T08:06:54 | |||||||||
ale-0328 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | auto-harness | https://github.com/neosigmaai/auto-harness | external | github.com | Bring-your-own-agent framework for self-improving agentic systems that mines failures from runs, optimizes the harness in response, and gates every change behind regression checks. | Bring-your-own-agent framework for self-improving agentic systems that mines failures from runs, optimizes the harness in response, and gates every change behind regression checks. | Bring-your-own-agent framework for self-improving agentic systems that mines failures from runs, optimizes the harness in response, and gates every change behind regression checks. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Bring-your-own-agent framework for self-improving agentic systems that mines failures from runs, optimizes the harness in response, and gates every change behind regression checks. | Use auto-harness to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (527 stars; 60 forks; MIT license; updated 2026-07-22); popularity is context, not proof of reliability. | medium | README.md | 965 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L965 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/neosigmaai/auto-harness | GitHub - neosigmaai/auto-harness: Bring your own agent and build a self-improving agentic system. Automatically mine failures, optimize the agent harness, and gate against regressions. · GitHub | Bring your own agent and build a self-improving agentic system. Automatically mine failures, optimize the agent harness, and gate against regressions. - neosigmaai/auto-harness | 2026-04-03 | 2026 | neosigmaai/auto-harness | GitHub | github-api | neosigmaai/auto-harness | 527 | 60 | MIT | 2026-04-03T21:18:14Z | 2026-07-22T16:36:56Z | 2026-07-29T08:06:54 | ||||||
ale-0329 | Verification And Feedback Gates | verification-and-feedback-gates | Docs | 📚 | OpenAI agent evals | https://developers.openai.com/api/docs/guides/agent-evals | external | developers.openai.com | Evaluation guidance for moving from traces to repeatable grading of agent workflows. | Evaluation guidance for moving from traces to repeatable grading of agent workflows. | Evaluation guidance for moving from traces to repeatable grading of agent workflows. | Evaluation data is used as the feedback signal for improving loop behavior. Evaluation guidance for moving from traces to repeatable grading of agent workflows. | Use OpenAI agent evals to measure progress and gate completion with repeatable evidence. | Primary official documentation from developers.openai.com; use it for current product or standard behavior. | high | README.md | 970 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L970 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | official-documentation | A | ok | https://developers.openai.com/api/docs/guides/agent-evals | Evaluate agent workflows | OpenAI API | Learn how to evaluate agent workflows with traces, graders, datasets, and evaluation runs on the OpenAI platform. | OpenAI Developers | html-meta | 2026-07-29T08:06:54 | |||||||||||||||
ale-0330 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Promptfoo OpenAI Agents provider | https://www.promptfoo.dev/docs/providers/openai-agents/ | external | www.promptfoo.dev | Testing and assertions for multi-turn agent workflows, tools, state, handoffs, sandboxes, and traces. | Testing and assertions for multi-turn agent workflows, tools, state, handoffs, sandboxes, and traces. | Testing and assertions for multi-turn agent workflows, tools, state, handoffs, sandboxes, and traces. | Execution isolation and permission boundaries are part of the design. Testing and assertions for multi-turn agent workflows, tools, state, handoffs, sandboxes, and traces. | Use Promptfoo OpenAI Agents provider to measure progress and gate completion with repeatable evidence. | Working implementation, framework, runtime, or repository; signal comes from usable code and ecosystem adoption. | high | README.md | 971 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L971 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;delegation;verification;state | builder;evaluator | harness | enabling | implementation | A | ok | https://www.promptfoo.dev/docs/providers/openai-agents/ | OpenAI Agents | Promptfoo | Test OpenAI Agents with tools, handoffs, sessions, sandbox workflows, and tracing in promptfoo. | promptfoo.dev | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0331 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Inspect AI | https://github.com/UKGovernmentBEIS/inspect_ai | external | github.com | UK AISI evaluation framework with solvers, scorers, sandboxing, tool use, MCP, and log viewing. | UK AISI evaluation framework with solvers, scorers, sandboxing, tool use, MCP, and log viewing. | UK AISI evaluation framework with solvers, scorers, sandboxing, tool use, MCP, and log viewing. | Evaluation data is used as the feedback signal for improving loop behavior. UK AISI evaluation framework with solvers, scorers, sandboxing, tool use, MCP, and log viewing. | Use Inspect AI to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (2,427 stars; 624 forks; MIT license; updated 2026-07-28); popularity is context, not proof of reliability. | medium | README.md | 972 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L972 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/UKGovernmentBEIS/inspect_ai | GitHub - UKGovernmentBEIS/inspect_ai: Inspect: A framework for large language model evaluations · GitHub | Inspect: A framework for large language model evaluations - UKGovernmentBEIS/inspect_ai | 2023-11-14 | 2023 | UKGovernmentBEIS/inspect_ai | GitHub | github-api | UKGovernmentBEIS/inspect_ai | 2427 | 624 | MIT | 2023-11-14T14:53:11Z | 2026-07-28T21:27:57Z | 2026-07-29T08:06:54 | ||||||
ale-0332 | Verification And Feedback Gates | verification-and-feedback-gates | Docs | 📚 | OpenTelemetry Semantic Conventions for Generative AI Systems | https://opentelemetry.io/docs/specs/semconv/gen-ai/ | external | opentelemetry.io | Portable tracing conventions for model calls, tool calls, and agent workflows. | Portable tracing conventions for model calls, tool calls, and agent workflows. | Portable tracing conventions for model calls, tool calls, and agent workflows. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Portable tracing conventions for model calls, tool calls, and agent workflows. | Use OpenTelemetry Semantic Conventions for Generative AI Systems to measure progress and gate completion with repeatable evidence. | Primary official documentation from opentelemetry.io; use it for current product or standard behavior. | high | README.md | 973 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L973 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace | builder;evaluator | harness | enabling | official-documentation | A | ok | https://opentelemetry.io/docs/specs/semconv/gen-ai/ | Moved: Generative AI semantic conventions | OpenTelemetry The OpenTelemetry Logo | Important GenAI semantic conventions have moved to the OpenTelemetry GenAI semantic conventions repository. This page has moved and is no longer maintained in this repository. | OpenTelemetry | html-meta | 2026-07-29T08:06:54 | |||||||||||||||
ale-0333 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | AgentOps | https://github.com/AgentOps-AI/agentops | external | github.com | Monitoring, replay, cost tracking, benchmarking, and tracing for agent sessions. | Monitoring, replay, cost tracking, benchmarking, and tracing for agent sessions. | Monitoring, replay, cost tracking, benchmarking, and tracing for agent sessions. | Durable execution and replay are treated as first-class loop infrastructure. Monitoring, replay, cost tracking, benchmarking, and tracing for agent sessions. | Use AgentOps to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (5,738 stars; 612 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 974 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L974 | Verify | verify | Gate progress with tests, evals, and evidence. | state;budget | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/AgentOps-AI/agentops | GitHub - AgentOps-AI/agentops: Python SDK for AI agent monitoring, LLM cost tracking, benchmarking, and more. Integrates with most LLMs and agent frameworks including CrewAI, Agno, OpenAI Agents SDK, Langchain, Autogen, AG2, and CamelAI · GitHub | Python SDK for AI agent monitoring, LLM cost tracking, benchmarking, and more. Integrates with most LLMs and agent frameworks including CrewAI, Agno, OpenAI Agents SDK, Langchain, Autogen, AG2, and CamelAI - AgentOps-AI/agentops | 2023-08-15 | 2023 | AgentOps-AI/agentops | GitHub | github-api | AgentOps-AI/agentops | 5738 | 612 | MIT | 2023-08-15T23:26:23Z | 2026-07-29T07:52:53Z | 2026-07-29T08:06:54 | ||||||
ale-0334 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Langfuse | https://github.com/langfuse/langfuse | external | github.com | Open-source LLM engineering platform with tracing, evaluations, and metrics that loops can read back as feedback signals. | Open-source LLM engineering platform with tracing, evaluations, and metrics that loops can read back as feedback signals. | Open-source LLM engineering platform with tracing, evaluations, and metrics that loops can read back as feedback signals. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Open-source LLM engineering platform with tracing, evaluations, and metrics that loops can read back as feedback signals. | Use Langfuse to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (32,064 stars; 3,430 forks; NOASSERTION license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 975 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L975 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/langfuse/langfuse | GitHub - langfuse/langfuse: 🪢 Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datasets. Integrates with OpenTelemetry, LangChain, OpenAI SDK, LiteLLM, and more. 🍊YC W23 · GitHub | 🪢 Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datasets. Integrates with OpenTelemetry, LangChain, OpenAI SDK, LiteLLM, and more. 🍊YC W23 - GitHub - langfuse/langfuse: 🪢 Open source AI engineering platform: LLM evals, observability, metrics, prompt management... | 2023-05-18 | 2023 | langfuse/langfuse | GitHub | github-api | langfuse/langfuse | 32064 | 3430 | NOASSERTION | 2023-05-18T17:47:09Z | 2026-07-29T07:57:04Z | 2026-07-29T08:06:54 | ||||||
ale-0335 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | LangSmith | https://www.langchain.com/langsmith | external | www.langchain.com | Tracing, evaluation, and monitoring platform for inspecting and grading agent runs across iterations. | Tracing, evaluation, and monitoring platform for inspecting and grading agent runs across iterations. | Tracing, evaluation, and monitoring platform for inspecting and grading agent runs across iterations. | Evaluation data is used as the feedback signal for improving loop behavior. Tracing, evaluation, and monitoring platform for inspecting and grading agent runs across iterations. | Use LangSmith to measure progress and gate completion with repeatable evidence. | Working implementation, framework, runtime, or repository; signal comes from usable code and ecosystem adoption. | high | README.md | 976 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L976 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | implementation | A | ok | https://www.langchain.com/langsmith/observability | LangSmith: Agent & LLM Observability Platform | Complete AI agent and LLM observability platform with tracing and real-time monitoring. Debug agents, find failures fast, and track costs and latency. | LangChain | domain-fallback | 2026-07-29T08:06:54 | |||||||||||||||
ale-0336 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Arize Phoenix | https://github.com/Arize-ai/phoenix | external | github.com | Open-source AI observability for tracing, evaluating, and debugging agent behavior from real runs. | Open-source AI observability for tracing, evaluating, and debugging agent behavior from real runs. | Open-source AI observability for tracing, evaluating, and debugging agent behavior from real runs. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Open-source AI observability for tracing, evaluating, and debugging agent behavior from real runs. | Use Arize Phoenix to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (10,792 stars; 1,016 forks; NOASSERTION license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 977 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L977 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/Arize-ai/phoenix | GitHub - Arize-ai/phoenix: AI Observability & Evaluation · GitHub | AI Observability & Evaluation. Contribute to Arize-ai/phoenix development by creating an account on GitHub. | 2022-11-09 | 2022 | Arize-ai/phoenix | GitHub | github-api | Arize-ai/phoenix | 10792 | 1016 | NOASSERTION | 2022-11-09T23:44:35Z | 2026-07-29T06:47:41Z | 2026-07-29T08:06:54 | ||||||
ale-0337 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Braintrust | https://www.braintrust.dev/ | external | www.braintrust.dev | Evaluation and observability platform with experiments, datasets, and CI integration for gating agent changes. | Evaluation and observability platform with experiments, datasets, and CI integration for gating agent changes. | Evaluation and observability platform with experiments, datasets, and CI integration for gating agent changes. | Evaluation data is used as the feedback signal for improving loop behavior. Evaluation and observability platform with experiments, datasets, and CI integration for gating agent changes. | Use Braintrust to measure progress and gate completion with repeatable evidence. | Working implementation, framework, runtime, or repository; signal comes from usable code and ecosystem adoption. | high | README.md | 978 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L978 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | implementation | A | ok | https://www.braintrust.dev/ | Braintrust - The AI observability platform for building quality AI products | Ship quality agents at scale. Braintrust is the AI observability platform for tracing production, running evals, and catching regressions before they reach users. | Braintrust | html-meta | 2026-07-29T08:06:54 | |||||||||||||||
ale-0338 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Weave | https://docs.wandb.ai/weave | external | docs.wandb.ai | Weights & Biases toolkit for tracing, evaluating, and monitoring agent applications over time. | Weights & Biases toolkit for tracing, evaluating, and monitoring agent applications over time. | Weights & Biases toolkit for tracing, evaluating, and monitoring agent applications over time. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Weights & Biases toolkit for tracing, evaluating, and monitoring agent applications over time. | Use Weave to measure progress and gate completion with repeatable evidence. | Working implementation, framework, runtime, or repository; signal comes from usable code and ecosystem adoption. | high | README.md | 979 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L979 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | implementation | A | ok | https://docs.wandb.ai/weave | W&B Weave - Weights & Biases Documentation | Track, test, and improve language model apps with W&B Weave | Weights & Biases Documentation | html-meta | 2026-07-29T08:06:54 | |||||||||||||||
ale-0339 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | agentops (boshu2) | https://github.com/boshu2/agentops | external | github.com | Independent verification layer for coding agents where a change only counts as done after a different model or a real test checks it, with the verdict recorded in the repo via a tamper-evident ledger. | Independent verification layer for coding agents where a change only counts as done after a different model or a real test checks it, with the verdict recorded in the repo via a tamper-evident ledger. | Independent verification layer for coding agents where a change only counts as done after a different model or a real test checks it, with the verdict recorded in the repo via a tamper-evident ledger. | Verification is promoted from a final check to a loop-control signal. Independent verification layer for coding agents where a change only counts as done after a different model or a real test checks it, with the verdict recorded in the repo via a tamper-evident ledger. | Use agentops (boshu2) to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (415 stars; 40 forks; Apache-2.0 license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 980 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L980 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;exit | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/boshu2/agentops | GitHub - boshu2/agentops: The operating loop a coding agent follows — and skills to orchestrate multi-agent systems. · GitHub | The operating loop a coding agent follows — and skills to orchestrate multi-agent systems. - boshu2/agentops | 2025-11-05 | 2025 | boshu2/agentops | GitHub | github-api | boshu2/agentops | 415 | 40 | Apache-2.0 | 2025-11-05T19:18:56Z | 2026-07-29T07:53:07Z | 2026-07-29T08:06:54 | ||||||
ale-0340 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | SkillSpec | https://github.com/modiqo/skillspec | external | github.com | CLI that makes agent skills followable, testable, and provable by converting prose skills into structured contracts, scoring follow-through risk, and generating execution traces of which steps ran, were skipped, and what evidence exists. | CLI that makes agent skills followable, testable, and provable by converting prose skills into structured contracts, scoring follow-through risk, and generating execution traces of which steps ran, were skipped, and what evidence exists. | CLI that makes agent skills followable, testable, and provable by converting prose skills into structured contracts, scoring follow-through risk, and generating execution traces of which steps ran, were skipped, and what evidence exists. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. CLI that makes agent skills followable, testable, and provable by converting prose skills into structured contracts, scoring follow-through risk, and generating execution traces of which steps ran, were skipped, and what evidence exists. | Use SkillSpec to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (909 stars; 61 forks; Apache-2.0 license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 981 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L981 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/modiqo/skillspec | GitHub - modiqo/skillspec: SkillSpec makes agent skills followable, testable, and provable with Doctor risk reports, guided imports, structured contracts, and alignment proof. · GitHub | SkillSpec makes agent skills followable, testable, and provable with Doctor risk reports, guided imports, structured contracts, and alignment proof. - modiqo/skillspec | 2026-06-19 | 2026 | modiqo/skillspec | GitHub | github-api | modiqo/skillspec | 909 | 61 | Apache-2.0 | 2026-06-19T23:42:55Z | 2026-07-29T04:06:02Z | 2026-07-29T08:06:54 | ||||||
ale-0341 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Shepherd | https://github.com/shepherd-agents/shepherd | external | github.com | Python runtime that records agent execution as reversible, Git-like traces so meta-agents or humans can observe, fork, replay, and revert any run before results touch files, with copy-on-write forking, roughly 95% cache reuse on replay, and syscall-level permission enforcement. | Python runtime that records agent execution as reversible, Git-like traces so meta-agents or humans can observe, fork, replay, and revert any run before results touch files, with copy-on-write forking, roughly 95% cache reuse on replay, and syscall-level permission enforcement. | Python runtime that records agent execution as reversible, Git-like traces so meta-agents or humans can observe, fork, replay, and revert any run before results touch files, with copy-on-write forking, roughly 95% cache reuse on replay, and syscall-level permission enforcement. | Durable execution and replay are treated as first-class loop infrastructure. Python runtime that records agent execution as reversible, Git-like traces so meta-agents or humans can observe, fork, replay, and revert any run before results touch files, with copy-on-write forking, roughly 95% cache reuse on replay, and sy... | Use Shepherd to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (1,587 stars; 124 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 982 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L982 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;state | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/shepherd-agents/shepherd | GitHub - shepherd-agents/shepherd: A runtime substrate that turns an agent's execution into a reversible, Git-like trace, so meta-agents can observe, fork, replay, and revert any run. Couples agent and environments in a copy-on-write fork ~5x faster than docker commit, with ~95% KV-cache reuse on replay. Framework buil... | A runtime substrate that turns an agent's execution into a reversible, Git-like trace, so meta-agents can observe, fork, replay, and revert any run. Couples agent and environments in a copy-on-write fork ~5x faster than docker commit, with ~95% KV-cache reuse on replay. Framework built for meta-agents to supervise, opt... | 2026-06-24 | 2026 | shepherd-agents/shepherd | GitHub | github-api | shepherd-agents/shepherd | 1587 | 124 | MIT | 2026-06-24T17:26:46Z | 2026-07-29T07:38:23Z | 2026-07-29T08:06:54 | ||||||
ale-0342 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | grill-for-unknowns | https://github.com/nicobailon/grill-for-unknowns | external | github.com | Portable SKILL.md agent skill that gates long-running subagent and coding-agent launches behind plan interrogation: it inspects the real territory (docs, source, tests, config) first, sorts what's known into facts, decisions, domain language, and unknowns across known/unknown quadrants, then emits a launch packet with ... | Portable SKILL.md agent skill that gates long-running subagent and coding-agent launches behind plan interrogation: it inspects the real territory (docs, source, tests, config) first, sorts what's known into facts, decisions, domain language, and unknowns across known/unknown quadrants, then emits a launch packet with ... | Portable SKILL.md agent skill that gates long-running subagent and coding-agent launches behind plan interrogation: it inspects the real territory (docs, source, tests, config) first, sorts what's known into facts, decisions, domain language, and unknowns across known/unknown quadrants, then emits a launch packet with ... | Verification is promoted from a final check to a loop-control signal. Portable SKILL.md agent skill that gates long-running subagent and coding-agent launches behind plan interrogation: it inspects the real territory (docs, source, tests, config) first, sorts what's known into facts, decisions, domain language, and unk... | Use grill-for-unknowns to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (190 stars; 7 forks; MIT license; updated 2026-07-26); popularity is context, not proof of reliability. | medium | README.md | 983 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L983 | Verify | verify | Gate progress with tests, evals, and evidence. | delegation;verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/nicobailon/grill-for-unknowns | GitHub - nicobailon/grill-for-unknowns: Agent skill for finding unknowns, grilling plans, and reaching shared understanding before implementation · GitHub | Agent skill for finding unknowns, grilling plans, and reaching shared understanding before implementation - nicobailon/grill-for-unknowns | 2026-07-09 | 2026 | nicobailon/grill-for-unknowns | GitHub | github-api | nicobailon/grill-for-unknowns | 190 | 7 | MIT | 2026-07-09T18:55:30Z | 2026-07-26T11:06:26Z | 2026-07-29T08:06:54 | ||||||
ale-0343 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Fable Harness | https://github.com/Miguok/fable-harness | external | github.com | Drop-in behavior protocol kit (hooks, a skill, and sub-agents auto-injected into every Claude Code session) enforcing a verify-first process: gather evidence before answering and verify changes before declaring done. | Drop-in behavior protocol kit (hooks, a skill, and sub-agents auto-injected into every Claude Code session) enforcing a verify-first process: gather evidence before answering and verify changes before declaring done. | Drop-in behavior protocol kit (hooks, a skill, and sub-agents auto-injected into every Claude Code session) enforcing a verify-first process: gather evidence before answering and verify changes before declaring done. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Drop-in behavior protocol kit (hooks, a skill, and sub-agents auto-injected into every Claude Code session) enforcing a verify-first process: gather evidence before answering and verify changes before declaring done. | Use Fable Harness to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (197 stars; 35 forks; MIT license; updated 2026-07-28); popularity is context, not proof of reliability. | medium | README.md | 984 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L984 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;exit | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/Miguok/fable-harness | GitHub - Miguok/fable-harness: Make Claude Code work like a disciplined engineer: OODA, multi-party adversarial review, tiered model routing, fail-then-pass — token-efficient by design (route heavy work to smaller models, isolate sub-agent context). Distilled from Fable to reinforce the Opus harness. · GitHub | Make Claude Code work like a disciplined engineer: OODA, multi-party adversarial review, tiered model routing, fail-then-pass — token-efficient by design (route heavy work to smaller models, isolate sub-agent context). Distilled from Fable to reinforce the Opus harness. - Miguok/fable-harness | 2026-07-05 | 2026 | Miguok/fable-harness | GitHub | github-api | Miguok/fable-harness | 197 | 35 | MIT | 2026-07-05T05:57:40Z | 2026-07-28T03:49:17Z | 2026-07-29T08:06:54 | ||||||
ale-0344 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Mindwalk | https://github.com/cosmtrek/mindwalk | external | github.com | Local visualization tool that replays Claude Code and Codex session logs as light moving across a 3D map of the repository, making long agent runs inspectable after the fact. | Local visualization tool that replays Claude Code and Codex session logs as light moving across a 3D map of the repository, making long agent runs inspectable after the fact. | Local visualization tool that replays Claude Code and Codex session logs as light moving across a 3D map of the repository, making long agent runs inspectable after the fact. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Local visualization tool that replays Claude Code and Codex session logs as light moving across a 3D map of the repository, making long agent runs inspectable after the fact. | Use Mindwalk to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (948 stars; 68 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 985 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L985 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/cosmtrek/mindwalk | GitHub - cosmtrek/mindwalk: A visualization tool that replays coding-agent sessions on a 3D map of your codebase. · GitHub | A visualization tool that replays coding-agent sessions on a 3D map of your codebase. - cosmtrek/mindwalk | 2026-07-09 | 2026 | cosmtrek/mindwalk | GitHub | github-api | cosmtrek/mindwalk | 948 | 68 | MIT | 2026-07-09T11:41:46Z | 2026-07-29T04:35:49Z | 2026-07-29T08:06:54 | ||||||
ale-0345 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Waggle | https://github.com/modiqo/waggle | external | github.com | MCP-native reference layer for agent handoffs: instead of pasting full context between agents, it passes a compact attributed, resolvable reference token that the receiving agent expands on demand. | MCP-native reference layer for agent handoffs: instead of pasting full context between agents, it passes a compact attributed, resolvable reference token that the receiving agent expands on demand. | MCP-native reference layer for agent handoffs: instead of pasting full context between agents, it passes a compact attributed, resolvable reference token that the receiving agent expands on demand. | Context is managed as durable loop state rather than a single prompt payload. MCP-native reference layer for agent handoffs: instead of pasting full context between agents, it passes a compact attributed, resolvable reference token that the receiving agent expands on demand. | Use Waggle to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (860 stars; 110 forks; Apache-2.0 license; updated 2026-07-28); popularity is context, not proof of reliability. | medium | README.md | 986 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L986 | 2026-07-15 | Verify | verify | Gate progress with tests, evals, and evidence. | context;delegation;budget | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/modiqo/waggle | GitHub - modiqo/waggle: Attributed, resolvable artifact references for agent handoffs — a ~30-byte token instead of pasted context. MCP-native; the reference layer for the agent-harness world. · GitHub | Attributed, resolvable artifact references for agent handoffs — a ~30-byte token instead of pasted context. MCP-native; the reference layer for the agent-harness world. - modiqo/waggle | 2026-07-08 | 2026 | modiqo/waggle | GitHub | github-api | modiqo/waggle | 860 | 110 | Apache-2.0 | 2026-07-08T04:33:09Z | 2026-07-28T00:13:00Z | 2026-07-29T08:06:54 | |||||
ale-0346 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Jacquard | https://github.com/jbwinters/jacquard-lang | external | github.com | Research language designed around the machine-writes, human-verifies contract, using effect-typed signatures so an agent's generated code carries checkable declarations of what it is allowed to touch. | Research language designed around the machine-writes, human-verifies contract, using effect-typed signatures so an agent's generated code carries checkable declarations of what it is allowed to touch. | Research language designed around the machine-writes, human-verifies contract, using effect-typed signatures so an agent's generated code carries checkable declarations of what it is allowed to touch. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Research language designed around the machine-writes, human-verifies contract, using effect-typed signatures so an agent's generated code carries checkable declarations of what it is allowed to touch. | Use Jacquard to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (109 stars; 3 forks; Apache-2.0 license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 987 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L987 | 2026-07-15 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/jbwinters/jacquard-lang | GitHub - jbwinters/jacquard-lang: Jacquard is a small programming language designed for a regime in which most code is written by machine-learning models and reviewed by people. · GitHub | Jacquard is a small programming language designed for a regime in which most code is written by machine-learning models and reviewed by people. - jbwinters/jacquard-lang | 2026-07-06 | 2026 | jbwinters/jacquard-lang | GitHub | github-api | jbwinters/jacquard-lang | 109 | 3 | Apache-2.0 | 2026-07-06T23:14:47Z | 2026-07-29T00:51:02Z | 2026-07-29T08:06:54 | |||||
ale-0347 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Agentic Verification of Software Systems | https://arxiv.org/abs/2511.17330 | external | arxiv.org | Pairs a coding agent with a theorem prover (AutoRocq) in a generate-and-validate loop, turning formal proof into the exit gate for trusted automatic programming. | Pairs a coding agent with a theorem prover (AutoRocq) in a generate-and-validate loop, turning formal proof into the exit gate for trusted automatic programming. | Pairs a coding agent with a theorem prover (AutoRocq) in a generate-and-validate loop, turning formal proof into the exit gate for trusted automatic programming. | Verification is promoted from a final check to a loop-control signal. Pairs a coding agent with a theorem prover (AutoRocq) in a generate-and-validate loop, turning formal proof into the exit gate for trusted automatic programming. | Use Agentic Verification of Software Systems to measure progress and gate completion with repeatable evidence. | Research source arXiv:2511.17330; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 992 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L992 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;exit | researcher;evaluator | harness | enabling | research-paper | A | ok | https://doi.org/10.1145/3808164 | [2511.17330] Agentic Verification of Software Systems | Automatically generated code is gaining traction recently, owing to the prevalence of Large Language Models (LLMs). Further, the AlphaProof initiative has demonstrated the possibility of using AI for general mathematical reasoning. Reasoning about computer programs (software) can be accomplished via general mathematica... | Haoxin Tu; Huan Zhao; Yahui Song; Mehtab Zafar; Ruijie Meng; Abhik Roychoudhury | 2026-06-30 | 2026 | Proceedings of the ACM on Software Engineering 3 (FSE) | Association for Computing Machinery | 10.1145/3808164 | Published in Proceedings of the ACM on Software Engineering 3 (FSE); the linked arXiv record remains available for open access. | cs.SE | ACM DOI record | 2511.17330 | 2026-07-29T08:06:54 | |||||||
ale-0348 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | A Trace-Based Assurance Framework for Agentic AI Orchestration: Contracts, Testing, and Governance | https://arxiv.org/abs/2603.18096 | external | arxiv.org | Treats execution traces as the assurance substrate, pairing machine-checkable contracts, testing, and governance so recurring agent orchestration stays verifiable and auditable. | Treats execution traces as the assurance substrate, pairing machine-checkable contracts, testing, and governance so recurring agent orchestration stays verifiable and auditable. | Treats execution traces as the assurance substrate, pairing machine-checkable contracts, testing, and governance so recurring agent orchestration stays verifiable and auditable. | Orchestration and control flow are made explicit and inspectable. Treats execution traces as the assurance substrate, pairing machine-checkable contracts, testing, and governance so recurring agent orchestration stays verifiable and auditable. | Use A Trace-Based Assurance Framework for Agentic AI Orchestration: Contracts, Testing, and Governance to measure progress and gate completion with repeatable evidence. | Research source arXiv:2603.18096; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 993 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L993 | Verify | verify | Gate progress with tests, evals, and evidence. | delegation;verification | researcher;evaluator | harness | enabling | research-paper | A | ok | https://doi.org/10.5220/0014840300004015 | [2603.18096] A Trace-Based Assurance Framework for Agentic AI Orchestration: Contracts, Testing, and Governance | In Agentic AI, Large Language Models (LLMs) are increasingly used in the orchestration layer to coordinate multiple agents and to interact with external services, retrieval components, and shared memory. In this setting, failures are not limited to incorrect final outputs. They also arise from long-horizon interaction,... | Ciprian Paduraru; Petru-Liviu Bouruc; Alin Stefanescu | 2026 | 2026 | Proceedings of the 21st International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE) | SCITEPRESS | 10.5220/0014840300004015 | Published in Proceedings of the 21st International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE); the linked arXiv record remains available for open access. | cs.MA | SCITEPRESS DOI record | 2603.18096 | 2026-07-29T08:06:54 | |||||||
ale-0349 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Self-Evolving Agents with Anytime-Valid Certificates | https://arxiv.org/abs/2607.00871 | external | arxiv.org | Confines self-modification to a small steering adapter around a frozen base model and gates each change with anytime-valid statistical tests that emit auditable certificates, reporting solve-count gains and logged regression prevention on a SWE-bench Verified subset. | Confines self-modification to a small steering adapter around a frozen base model and gates each change with anytime-valid statistical tests that emit auditable certificates, reporting solve-count gains and logged regression prevention on a SWE-bench Verified subset. | Confines self-modification to a small steering adapter around a frozen base model and gates each change with anytime-valid statistical tests that emit auditable certificates, reporting solve-count gains and logged regression prevention on a SWE-bench Verified subset. | Verification is promoted from a final check to a loop-control signal. Confines self-modification to a small steering adapter around a frozen base model and gates each change with anytime-valid statistical tests that emit auditable certificates, reporting solve-count gains and logged regression prevention on a SWE-bench... | Use Self-Evolving Agents with Anytime-Valid Certificates to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.00871; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 994 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L994 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.00871 | [2607.00871] Self-Evolving Agents with Anytime-Valid Certificates | Self-evolving agents violate the assumption behind most learning-theoretic guarantees: the data, evaluator, components, and hypothesis space are produced by the policy being updated. We present \textbf{SEA}, an architecture that confines self-modification to a small steering adapter and a versioned harness around a \em... | Biswa Sengupta | 2026-07-01 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.00871 | 2026-07-29T08:06:54 | |||||||||
ale-0350 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Delayed Verification Destabilizes Multi-Agent LLM Belief | https://arxiv.org/abs/2606.27409 | external | arxiv.org | Models verifier-corrector loops in multi-agent LLM systems as delayed consensus, deriving a stability threshold where verification that is too strong or too late turns factual consensus into oscillation, plus a greedy corrector-placement algorithm validated on five open models. | Models verifier-corrector loops in multi-agent LLM systems as delayed consensus, deriving a stability threshold where verification that is too strong or too late turns factual consensus into oscillation, plus a greedy corrector-placement algorithm validated on five open models. | Models verifier-corrector loops in multi-agent LLM systems as delayed consensus, deriving a stability threshold where verification that is too strong or too late turns factual consensus into oscillation, plus a greedy corrector-placement algorithm validated on five open models. | Verification is promoted from a final check to a loop-control signal. Models verifier-corrector loops in multi-agent LLM systems as delayed consensus, deriving a stability threshold where verification that is too strong or too late turns factual consensus into oscillation, plus a greedy corrector-placement algorithm va... | Use Delayed Verification Destabilizes Multi-Agent LLM Belief to measure progress and gate completion with repeatable evidence. | Research source arXiv:2606.27409; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 995 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L995 | Verify | verify | Gate progress with tests, evals, and evidence. | delegation;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2606.27409 | [2606.27409] Delayed Verification Destabilizes Multi-Agent LLM Belief: Instability Thresholds and Optimal Corrector Placement | Multi-agent large language model (LLM) systems often rely on verifier and critic agents to suppress hallucinations, but verification is delayed. During this delay, false claims can propagate through the agent network. We model this process as delayed consensus on a graph with grounded corrector nodes. Spectral decompos... | Igor Itkin | 2026-06-25 | 2026 | arXiv | arXiv | 20 pages, 5 figures, 1 table. Code and data: https://github.com/YehudaItkin/delayed-verification-llm | cs.MA | arxiv-api | 2606.27409 | 2026-07-29T08:06:54 | ||||||||
ale-0351 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory | https://arxiv.org/abs/2606.06523 | external | arxiv.org | Models agent workflows and trajectories in Lean 4 dependent types so semantic consistency is machine-checked rather than judged by an LLM, with verification-passing workflows outperforming failing ones by an average of 11.94% on software-engineering benchmarks. | Models agent workflows and trajectories in Lean 4 dependent types so semantic consistency is machine-checked rather than judged by an LLM, with verification-passing workflows outperforming failing ones by an average of 11.94% on software-engineering benchmarks. | Models agent workflows and trajectories in Lean 4 dependent types so semantic consistency is machine-checked rather than judged by an LLM, with verification-passing workflows outperforming failing ones by an average of 11.94% on software-engineering benchmarks. | Verification is promoted from a final check to a loop-control signal. Models agent workflows and trajectories in Lean 4 dependent types so semantic consistency is machine-checked rather than judged by an LLM, with verification-passing workflows outperforming failing ones by an average of 11.94% on software-engineering ... | Use Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory to measure progress and gate completion with repeatable evidence. | Research source arXiv:2606.06523; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 996 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L996 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2606.06523 | [2606.06523] Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory | Equipping Large Language Models (LLMs) to execute reliable multi-step workflows has become a central challenge in artificial intelligence. Despite recent advances in LLMs' agentic capabilities, most agent systems still lack formal methods for specifying, verifying, and debugging their workflow and execution trajectorie... | Ruida Wang; Jerry Huang; Pengcheng Wang; Xuanqing Liu; Luyang Kong; Tong Zhang | 2026-06-02 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2606.06523 | 2026-07-29T08:06:54 | |||||||||
ale-0352 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Regimes: An Auditable, Held-Out-Gated Improvement Loop | https://arxiv.org/abs/2606.10241 | external | arxiv.org | Event-sourced agent runtime whose self-improvement loop gates every proposed repair behind static checks, sandbox execution, and held-out evaluation before adoption, keeping the full decision trail replayable. | Event-sourced agent runtime whose self-improvement loop gates every proposed repair behind static checks, sandbox execution, and held-out evaluation before adoption, keeping the full decision trail replayable. | Event-sourced agent runtime whose self-improvement loop gates every proposed repair behind static checks, sandbox execution, and held-out evaluation before adoption, keeping the full decision trail replayable. | Evaluation data is used as the feedback signal for improving loop behavior. Event-sourced agent runtime whose self-improvement loop gates every proposed repair behind static checks, sandbox execution, and held-out evaluation before adoption, keeping the full decision trail replayable. | Use Regimes: An Auditable, Held-Out-Gated Improvement Loop to measure progress and gate completion with repeatable evidence. | Research source arXiv:2606.10241; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 997 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L997 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2606.10241 | [2606.10241] Regimes: An Auditable, Held-Out-Gated Improvement Loop Demonstrated on LongMemEval with ActiveGraph | Autonomous improvement loops are hard to trust because the improvement process is usually external scaffolding bolted onto the agent: failures go unlogged, diagnoses cannot be replayed, and promote-or-discard decisions land in a side database rather than the agent's own history. We show that an event-sourced agent runt... | Yohei Nakajima | 2026-06-08 | 2026 | arXiv | arXiv | 30 pages, 5 figures. Code and committed runs: https://github.com/yoheinakajima/regimes | cs.AI | arxiv-api | 2606.10241 | 2026-07-29T08:06:54 | ||||||||
ale-0353 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents | https://arxiv.org/abs/2605.22608 | external | arxiv.org | Automated evaluation framework from IBM Research that grades agent behavior at system, trace, and node granularity without predefined error taxonomies, producing feedback aligned with human-annotated errors and predictive of task success. | Automated evaluation framework from IBM Research that grades agent behavior at system, trace, and node granularity without predefined error taxonomies, producing feedback aligned with human-annotated errors and predictive of task success. | Automated evaluation framework from IBM Research that grades agent behavior at system, trace, and node granularity without predefined error taxonomies, producing feedback aligned with human-annotated errors and predictive of task success. | Evaluation data is used as the feedback signal for improving loop behavior. Automated evaluation framework from IBM Research that grades agent behavior at system, trace, and node granularity without predefined error taxonomies, producing feedback aligned with human-annotated errors and predictive of task success. | Use Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2605.22608; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 998 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L998 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | researcher;evaluator | harness | enabling | research-paper | A | ok | https://aclanthology.org/2026.acl-demo.74/ | [2605.22608] Agentic CLEAR: Automating Multi-Level Evaluation of LLM Agents | Agentic systems are becoming more capable: agents define strategies, take actions, and interact with different environments. This autonomy poses serious challenges for overseeing and assessing agent behavior. Most current tools are limited, focusing on observability with basic evaluation capabilities or imposing static... | Asaf Yehudai; Lilach Eden; Michal Shmueli-Scheuer | 2026 | 2026 | Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics: System Demonstrations (ACL) | Association for Computational Linguistics | 10.18653/v1/2026.acl-demo.74 | Published in Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics: System Demonstrations (ACL); the linked arXiv record remains available for open access. | cs.CL | ACL Anthology and DOI records | 2605.22608 | 2026-07-29T08:06:54 | |||||||
ale-0354 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Diagnosis-Driven Automatic Repair for Agentic Workflow via Symbolic Inference | https://arxiv.org/abs/2607.02882 | external | arxiv.org | FlowFixer converts runs of platform-built agentic workflows (Dify, Coze, n8n) into symbolic traces, infers correctness specs and node dependencies to localize root-cause failures, and generates targeted repairs at a 71.3% success rate. | FlowFixer converts runs of platform-built agentic workflows (Dify, Coze, n8n) into symbolic traces, infers correctness specs and node dependencies to localize root-cause failures, and generates targeted repairs at a 71.3% success rate. | FlowFixer converts runs of platform-built agentic workflows (Dify, Coze, n8n) into symbolic traces, infers correctness specs and node dependencies to localize root-cause failures, and generates targeted repairs at a 71.3% success rate. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. FlowFixer converts runs of platform-built agentic workflows (Dify, Coze, n8n) into symbolic traces, infers correctness specs and node dependencies to localize root-cause failures, and generates targeted repairs at a 71.3% success rate. | Use Diagnosis-Driven Automatic Repair for Agentic Workflow via Symbolic Inference to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.02882; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 999 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L999 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.02882 | [2607.02882] Diagnosis-Driven Automatic Repair for Agentic Workflow via Symbolic Inference | Platform-orchestrated agentic workflows have become a popular paradigm for developing LLM-based applications. However, their reliability remains a major challenge due to the uncertainty of LLM outputs, complex inter-node dependencies, and heterogeneous tool interactions. Existing agentic workflow optimization and agent... | Xuyan Ma; Yawen Wang; Junjie Wang; Xiaofei Xie; Boyu Wu; Mingyang Li; Dandan Wang; Qing Wang | 2026-07-03 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.02882 | 2026-07-29T08:06:54 | |||||||||
ale-0355 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use | https://arxiv.org/abs/2607.01874 | external | arxiv.org | Self-evolving rubric framework that scores agent trajectories on skill selection, following, composition, and reflection, exposing failures that pass/fail outcome checks miss and beating outcome-only filtering as a training signal. | Self-evolving rubric framework that scores agent trajectories on skill selection, following, composition, and reflection, exposing failures that pass/fail outcome checks miss and beating outcome-only filtering as a training signal. | Self-evolving rubric framework that scores agent trajectories on skill selection, following, composition, and reflection, exposing failures that pass/fail outcome checks miss and beating outcome-only filtering as a training signal. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Self-evolving rubric framework that scores agent trajectories on skill selection, following, composition, and reflection, exposing failures that pass/fail outcome checks miss and beating outcome-only filtering as a training signal. | Use SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.01874; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,000 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1000 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.01874 | [2607.01874] SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use | Skills are becoming a reusable operational layer for LLM agents, encoding SOPs, domain rules, tool workflows, scripts, and validation routines. In realistic skill repositories, overlapping skills make reliable skill-use difficult. Final verifier success is too coarse for both evaluation and training, since an agent may... | Jiayin Zhu; Kelong Mao; Yudong Guo; Dengbo He; Sulong Xu; Simiu Gu; Yutao Yue | 2026-07-02 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.01874 | 2026-07-29T08:06:54 | |||||||||
ale-0356 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | SWE-Doctor: Guiding Software Engineering Agents with Runtime Diagnosis from Bug Reproduction Tests | https://arxiv.org/abs/2607.00990 | external | arxiv.org | Shows that naively feeding bug-reproduction tests to software-engineering agents can mislead them, and instead pipes runtime diagnosis from multi-faceted reproduction tests into patch generation, reaching 75.7% on SWE-bench Verified. | Shows that naively feeding bug-reproduction tests to software-engineering agents can mislead them, and instead pipes runtime diagnosis from multi-faceted reproduction tests into patch generation, reaching 75.7% on SWE-bench Verified. | Shows that naively feeding bug-reproduction tests to software-engineering agents can mislead them, and instead pipes runtime diagnosis from multi-faceted reproduction tests into patch generation, reaching 75.7% on SWE-bench Verified. | Verification is promoted from a final check to a loop-control signal. Shows that naively feeding bug-reproduction tests to software-engineering agents can mislead them, and instead pipes runtime diagnosis from multi-faceted reproduction tests into patch generation, reaching 75.7% on SWE-bench Verified. | Use SWE-Doctor: Guiding Software Engineering Agents with Runtime Diagnosis from Bug Reproduction Tests to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.00990; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,001 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1001 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.00990 | [2607.00990] SWE-Doctor: Guiding Software Engineering Agents with Runtime Diagnosis from Multi-Faceted Bug Reproduction Tests | Large language model (LLM)-based software engineering agents are increasingly developed to resolve software issues by generating patches from issue reports and code repositories. Bug reproduction tests (BRTs) are an important building block for such agents and have been shown useful for patch validation. However, it re... | Yaoqi Guo; Yang Liu; Jie M. Zhang; Yun Ma; Yiling Lou; Zhenpeng Chen | 2026-07-01 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.00990 | 2026-07-29T08:06:54 | |||||||||
ale-0357 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | AgentTether: Graph-Guided Diagnosis and Runtime Intervention for Reliable LLM Agent Operation | https://arxiv.org/abs/2607.06273 | external | arxiv.org | Runtime repair layer that abstracts agent runs into a dependency-aware critical-transition graph, localizes failure-critical subtrajectories after a run, and guides recovery on re-execution without modifying the underlying agent. | Runtime repair layer that abstracts agent runs into a dependency-aware critical-transition graph, localizes failure-critical subtrajectories after a run, and guides recovery on re-execution without modifying the underlying agent. | Runtime repair layer that abstracts agent runs into a dependency-aware critical-transition graph, localizes failure-critical subtrajectories after a run, and guides recovery on re-execution without modifying the underlying agent. | Control flow is represented as an inspectable graph rather than an opaque prompt loop. Runtime repair layer that abstracts agent runs into a dependency-aware critical-transition graph, localizes failure-critical subtrajectories after a run, and guides recovery on re-execution without modifying the underlying agent. | Use AgentTether: Graph-Guided Diagnosis and Runtime Intervention for Reliable LLM Agent Operation to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.06273; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,002 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1002 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.06273 | [2607.06273] AgentTether: Graph-Guided Diagnosis and Runtime Intervention for Reliable LLM Agent Operation | Large language model (LLM) agents are increasingly used for multi-step, stateful tool-use tasks, yet production reliability remains limited. Unlike static software repair, agent repair must recover dynamic trajectories whose early decisions can propagate into later errors and external state changes. Existing automatic ... | Chenyu Zhao; Shenglin Zhang; Wenwei Gu; Yongqian Sun; Dan Pei; Chetan Bansal; Saravan Rajmohan; Minghua Ma | 2026-07-07 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.06273 | 2026-07-29T08:06:54 | |||||||||
ale-0358 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | SWE-Review: Closing the Loop on Issue Resolution with Agentic Code Review | https://arxiv.org/abs/2607.06065 | external | arxiv.org | Replaces one-shot PR generation with a generate-review-revise loop in which a reviewer agent explores the repository, accepts or rejects the PR, and feeds structured feedback into revision, with an accompanying benchmark and trajectory dataset. | Replaces one-shot PR generation with a generate-review-revise loop in which a reviewer agent explores the repository, accepts or rejects the PR, and feeds structured feedback into revision, with an accompanying benchmark and trajectory dataset. | Replaces one-shot PR generation with a generate-review-revise loop in which a reviewer agent explores the repository, accepts or rejects the PR, and feeds structured feedback into revision, with an accompanying benchmark and trajectory dataset. | Packages the evidence as queryable CSV and JSONL rather than only a rendered page. Replaces one-shot PR generation with a generate-review-revise loop in which a reviewer agent explores the repository, accepts or rejects the PR, and feeds structured feedback into revision, with an accompanying benchmark and trajectory d... | Use SWE-Review: Closing the Loop on Issue Resolution with Agentic Code Review to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.06065; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,003 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1003 | Verify | verify | Gate progress with tests, evals, and evidence. | intake;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.06065 | [2607.06065] SWE-Review: Closing the Loop on Issue Resolution with Agentic Code Review | Coding agents increasingly generate pull requests (PRs) for real-world software issues, yet one-shot PR generation remains open-loop: the PR is proposed without systematic review, diagnosis, or revision. We introduce \textbf{SWE-Review}, a framework for closing this loop with agentic code review. Given an issue and an ... | Ruoyu Wang; Jierun Chen; Shaowei Wang; Chaofan Tao; Sidi Yang; Yuxin Jiang; Kim-Hui Yap; Lifeng Shang; Xiaohui Li; Haoli Bai | 2026-07-07 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.06065 | 2026-07-29T08:06:54 | |||||||||
ale-0359 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode | https://arxiv.org/abs/2607.07405 | external | arxiv.org | Finds that 78% of observed agent failures in a tau^2-bench domain are silent wrong-state failures invisible to both the tool and the agent's self-report, and that deterministic read-only pre-execution gates in the loop recover them. | Finds that 78% of observed agent failures in a tau^2-bench domain are silent wrong-state failures invisible to both the tool and the agent's self-report, and that deterministic read-only pre-execution gates in the loop recover them. | Finds that 78% of observed agent failures in a tau^2-bench domain are silent wrong-state failures invisible to both the tool and the agent's self-report, and that deterministic read-only pre-execution gates in the loop recover them. | State persistence is explicit enough for repeated runs and handoff. Finds that 78% of observed agent failures in a tau^2-bench domain are silent wrong-state failures invisible to both the tool and the agent's self-report, and that deterministic read-only pre-execution gates in the loop recover them. | Use Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07405; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,004 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1004 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification;state | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.07405 | [2607.07405] Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents | Tool-using LLM agents can violate the very policies they are deployed to enforce while appearing to complete the task successfully. In policy-permissive environments, a tool may execute any well-formed call even when the corresponding state transition is forbidden by domain policy. The result is a silent wrong state (a... | Vikas Reddy; Sumanth Reddy Challaram; Abhishek Basu | 2026-07-08 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.07405 | 2026-07-29T08:06:54 | |||||||||
ale-0360 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Harnessing Code Agents for Automatic Software Verification | https://arxiv.org/abs/2607.06341 | external | arxiv.org | Wraps a general code agent in a verification harness and lets it run until every targeted Coq lemma is proved, beating fixed human-designed proof strategies and reaching full lemma coverage with no expert intervention. | Wraps a general code agent in a verification harness and lets it run until every targeted Coq lemma is proved, beating fixed human-designed proof strategies and reaching full lemma coverage with no expert intervention. | Wraps a general code agent in a verification harness and lets it run until every targeted Coq lemma is proved, beating fixed human-designed proof strategies and reaching full lemma coverage with no expert intervention. | Verification is promoted from a final check to a loop-control signal. Wraps a general code agent in a verification harness and lets it run until every targeted Coq lemma is proved, beating fixed human-designed proof strategies and reaching full lemma coverage with no expert intervention. | Use Harnessing Code Agents for Automatic Software Verification to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.06341; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,005 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1005 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.06341 | [2607.06341] Harnessing Code Agents for Automatic Software Verification | Formal verification offers the strongest guarantee of software correctness, but it does not scale: the proofs demanded by interactive theorem provers such as Coq require enormous expert effort. Large language models (LLMs) promise to generate these proofs automatically, yet existing approaches wire a fixed, human-desig... | Shuangxiang Kan; Shuanglong Kan; Sebastian Ertel | 2026-07-07 | 2026 | arXiv | arXiv | cs.FL | arxiv-api | 2607.06341 | 2026-07-29T08:06:54 | |||||||||
ale-0361 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | LLM-as-a-Verifier: A General-Purpose Verification Framework | https://arxiv.org/abs/2607.05391 | external | arxiv.org | Treats verification as a scaling axis and builds a training-free framework that computes continuous scores from token logits for fine-grained agentic feedback, scaled via score granularity, repeated evaluation, and criteria decomposition. | Treats verification as a scaling axis and builds a training-free framework that computes continuous scores from token logits for fine-grained agentic feedback, scaled via score granularity, repeated evaluation, and criteria decomposition. | Treats verification as a scaling axis and builds a training-free framework that computes continuous scores from token logits for fine-grained agentic feedback, scaled via score granularity, repeated evaluation, and criteria decomposition. | Verification is promoted from a final check to a loop-control signal. Treats verification as a scaling axis and builds a training-free framework that computes continuous scores from token logits for fine-grained agentic feedback, scaled via score granularity, repeated evaluation, and criteria decomposition. | Use LLM-as-a-Verifier: A General-Purpose Verification Framework to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.05391; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,006 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1006 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;budget | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.05391 | [2607.05391] LLM-as-a-Verifier: A General-Purpose Verification Framework | Scaling pre-training, post-training, and test-time compute have become the central paradigms for improving the capabilities of LLMs. In this work, we identify verification, the ability to determine the correctness of a solution, as a new scaling axis. To unlock this and demonstrate its effectiveness, we introduce LLM-a... | Jacky Kwok; Shulu Li; Pranav Atreya; Yuejiang Liu; Yixing Jiang; Chelsea Finn; Marco Pavone; Ion Stoica; Azalia Mirhoseini | 2026-07-06 | 2026 | arXiv | arXiv | Code: https://github.com/llm-as-a-verifier/llm-as-a-verifier Website: https://llm-as-a-verifier.com | cs.AI | arxiv-api | 2607.05391 | 2026-07-29T08:06:54 | ||||||||
ale-0362 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents | https://arxiv.org/abs/2607.08028 | external | arxiv.org | Moves deterministic agent behavior out of prompts into code, schemas, and behavior contracts, wrapping validation around a replaceable model boundary so enterprise agents remain auditable and safe across model substitutions. | Moves deterministic agent behavior out of prompts into code, schemas, and behavior contracts, wrapping validation around a replaceable model boundary so enterprise agents remain auditable and safe across model substitutions. | Moves deterministic agent behavior out of prompts into code, schemas, and behavior contracts, wrapping validation around a replaceable model boundary so enterprise agents remain auditable and safe across model substitutions. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Moves deterministic agent behavior out of prompts into code, schemas, and behavior contracts, wrapping validation around a replaceable model boundary so enterprise agents remain auditable and safe across model substitutions. | Use From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.08028; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,007 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1007 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.08028 | [2607.08028] From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents | Enterprise large language model (LLM) applications often begin as prototypes whose behavior is carried by prompts and retrieval context. Productization adds requirements for source boundaries, entity routing, answer contracts, and reproducible traces. We present a harness-engineering approach that reconstructs this pat... | Joongho Ahn; Moonsoo Kim | 2026-07-09 | 2026 | arXiv | arXiv | 32 pages, 6 figures, 16 tables. Reference implementation and evaluation artifacts: https://github.com/hammerbaki/enterprise-llm-agent-harness (archived at https://doi.org/10.5281/zenodo.21269426) | cs.AI | arxiv-api | 2607.08028 | 2026-07-29T08:06:54 | ||||||||
ale-0363 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization | https://arxiv.org/abs/2607.07702 | external | arxiv.org | STRACE structures redundant, heterogeneous agent execution traces by mining batch-level failure patterns and performing causal localization over a textual dependency graph, handing root causes rather than noisy trajectories to the reflection-based optimizer and lifting success on a formal verification task from 42.5% t... | STRACE structures redundant, heterogeneous agent execution traces by mining batch-level failure patterns and performing causal localization over a textual dependency graph, handing root causes rather than noisy trajectories to the reflection-based optimizer and lifting success on a formal verification task from 42.5% t... | STRACE structures redundant, heterogeneous agent execution traces by mining batch-level failure patterns and performing causal localization over a textual dependency graph, handing root causes rather than noisy trajectories to the reflection-based optimizer and lifting success on a formal verification task from 42.5% t... | Control flow is represented as an inspectable graph rather than an opaque prompt loop. STRACE structures redundant, heterogeneous agent execution traces by mining batch-level failure patterns and performing causal localization over a textual dependency graph, handing root causes rather than noisy trajectories to the re... | Use From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07702; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,008 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1008 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.07702 | [2607.07702] From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization | The optimization of long-horizon agents increasingly relies on reflection-based mechanisms, where a large language model (LLM) acts as an optimizer to diagnose agent failures and improve agent policies. However, real execution traces are difficult to use directly for optimization: large trace collections are often redu... | Ying Chang; Jiahang Xu; Xuan Feng; Chenyuan Yang; Peng Cheng; Yuqing Yang | 2026-07-08 | 2026 | arXiv | arXiv | cs.CL | arxiv-api | 2607.07702 | 2026-07-29T08:06:54 | |||||||||
ale-0364 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems | https://arxiv.org/abs/2607.07989 | external | arxiv.org | AgentLocate attributes failures in LLM multi-agent trajectories to both the responsible agent and the earliest decisive step, pairing LLM-based evaluation with independent assessor verification and confidence-weighted aggregation to outperform prior attribution methods on two benchmarks, the diagnose side of the verify... | AgentLocate attributes failures in LLM multi-agent trajectories to both the responsible agent and the earliest decisive step, pairing LLM-based evaluation with independent assessor verification and confidence-weighted aggregation to outperform prior attribution methods on two benchmarks, the diagnose side of the verify... | AgentLocate attributes failures in LLM multi-agent trajectories to both the responsible agent and the earliest decisive step, pairing LLM-based evaluation with independent assessor verification and confidence-weighted aggregation to outperform prior attribution methods on two benchmarks, the diagnose side of the verify... | Verification is promoted from a final check to a loop-control signal. AgentLocate attributes failures in LLM multi-agent trajectories to both the responsible agent and the earliest decisive step, pairing LLM-based evaluation with independent assessor verification and confidence-weighted aggregation to outperform prior ... | Use Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07989; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,009 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1009 | Verify | verify | Gate progress with tests, evals, and evidence. | delegation;verification | researcher;evaluator | harness | enabling | research-paper | A | ok | https://arxiv.org/abs/2607.07989 | [2607.07989] Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems | Large language model (LLM) based multi-agent systems enable complex problem solving through coordinated reasoning and action, but their distributed structure also introduces new challenges in diagnosing system-level failures. When an execution fails, identifying which agent is responsible and at what point the trajecto... | Yufei Xia; Anjun Gao; Yueyang Quan; Zhuqing Liu; Minghong Fang | 2026 | 2026 | Conference on Language Modeling (COLM) | Conference on Language Modeling | Accepted at Conference on Language Modeling (COLM); the linked arXiv record is the available paper version. | cs.CR | Official COLM accepted-papers list and current arXiv note | 2607.07989 | 2026-07-29T08:06:54 | ||||||||
ale-0365 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | 3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse | https://arxiv.org/abs/2607.07980 | external | arxiv.org | Builds a causal theory of 26 constructs and 67 relationships from 3,100 coded practitioner documents on how AI-authored pull requests reshape code review, arguing review is the control point through which a coding agent's effect on software is decided and that outcomes hinge on team expertise and review process structu... | Builds a causal theory of 26 constructs and 67 relationships from 3,100 coded practitioner documents on how AI-authored pull requests reshape code review, arguing review is the control point through which a coding agent's effect on software is decided and that outcomes hinge on team expertise and review process structu... | Builds a causal theory of 26 constructs and 67 relationships from 3,100 coded practitioner documents on how AI-authored pull requests reshape code review, arguing review is the control point through which a coding agent's effect on software is decided and that outcomes hinge on team expertise and review process structu... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Builds a causal theory of 26 constructs and 67 relationships from 3,100 coded practitioner documents on how AI-authored pull requests reshape code review, arguing review is the control point through which a coding agent's effect on software ... | Use 3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07980; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,010 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1010 | Verify | verify | Gate progress with tests, evals, and evidence. | context | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.07980 | [2607.07980] 3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse | Coding agents now author entire pull requests, and practitioners sharply disagree about what this does to code review: whether it becomes the bottleneck, whether human review is still necessary, and whether it quietly erodes the understanding that it once built. Repository-mining studies measure surface trends but seld... | Shyam Agarwal; Courtney Miller; Christian Kästner; Bogdan Vasilescu | 2026-07-08 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.07980 | 2026-07-29T08:06:54 | |||||||||
ale-0366 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring | https://arxiv.org/abs/2607.08066 | external | arxiv.org | Stress-tests chain-of-thought monitoring as an in-loop safety gate: adversarial agents arguing for policy-violating proposals turn the scratchpad into a persuasion channel, with monitor access to the agent's reasoning increasing approval of harmful actions by 9.5% on average, while pairing monitor and fact-checker from... | Stress-tests chain-of-thought monitoring as an in-loop safety gate: adversarial agents arguing for policy-violating proposals turn the scratchpad into a persuasion channel, with monitor access to the agent's reasoning increasing approval of harmful actions by 9.5% on average, while pairing monitor and fact-checker from... | Stress-tests chain-of-thought monitoring as an in-loop safety gate: adversarial agents arguing for policy-violating proposals turn the scratchpad into a persuasion channel, with monitor access to the agent's reasoning increasing approval of harmful actions by 9.5% on average, while pairing monitor and fact-checker from... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Stress-tests chain-of-thought monitoring as an in-loop safety gate: adversarial agents arguing for policy-violating proposals turn the scratchpad into a persuasion channel, with monitor access to the agent's reasoning increasing approval of ... | Use Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.08066; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,011 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1011 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.08066 | [2607.08066] Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring | Chain-of-thought (CoT) monitoring is a promising safety mechanism for AI agents, based on the premise that visible reasoning traces can surface misaligned or deceptive behavior. While effective in standard scenarios, recent work highlights that LLMs remain vulnerable to persuasion-based jailbreaks, where natural-langua... | Jennifer Za; Julija Bainiaksina; Nikita Ostrovsky; Tanush Chopra; Victoria Krakovna | 2026-07-09 | 2026 | arXiv | arXiv | 25 pages, 10 figures | cs.AI | arxiv-api | 2607.08066 | 2026-07-29T08:06:54 | ||||||||
ale-0367 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Physics-Audited Agentic Discovery in Scientific Machine Learning | https://arxiv.org/abs/2607.07379 | external | arxiv.org | Verification-first workflow (PA-SciML) for agentic model discovery in scientific ML: fixes the scoring evaluator before search, derives machine-checkable physics requirements (boundary conditions, superposition, stiffness scaling, causality), audits every trained candidate's predicted fields against them, and separatel... | Verification-first workflow (PA-SciML) for agentic model discovery in scientific ML: fixes the scoring evaluator before search, derives machine-checkable physics requirements (boundary conditions, superposition, stiffness scaling, causality), audits every trained candidate's predicted fields against them, and separatel... | Verification-first workflow (PA-SciML) for agentic model discovery in scientific ML: fixes the scoring evaluator before search, derives machine-checkable physics requirements (boundary conditions, superposition, stiffness scaling, causality), audits every trained candidate's predicted fields against them, and separatel... | Verification is promoted from a final check to a loop-control signal. Verification-first workflow (PA-SciML) for agentic model discovery in scientific ML: fixes the scoring evaluator before search, derives machine-checkable physics requirements (boundary conditions, superposition, stiffness scaling, causality), audits ... | Use Physics-Audited Agentic Discovery in Scientific Machine Learning to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07379; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,012 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1012 | Verify | verify | Gate progress with tests, evals, and evidence. | intake;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.07379 | [2607.07379] Physics-Audited Agentic Discovery in Scientific Machine Learning | In agentic scientific machine learning (SciML), large language model (LLM) agents can discover surrogate models and select one by an automated score, typically an error metric. A low error, however, does not establish that the predicted fields satisfy the physics that matter for mechanics, such as boundary conditions, ... | Diab W. Abueidda; Bilal Ahmed; Panos Pantidis; Mostafa E. Mobasher | 2026-07-08 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.07379 | 2026-07-29T08:06:54 | |||||||||
ale-0368 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Bug Report Specification Refinement with Trajectory Guidance for Automated Program Repair | https://arxiv.org/abs/2607.07882 | external | arxiv.org | TrajSpec runs a trajectory-collection agent over the pre-fix repository and mines the unverified trajectory for specification evidence, refining vague bug reports into structured specifications that guide automated program-repair loops. | TrajSpec runs a trajectory-collection agent over the pre-fix repository and mines the unverified trajectory for specification evidence, refining vague bug reports into structured specifications that guide automated program-repair loops. | TrajSpec runs a trajectory-collection agent over the pre-fix repository and mines the unverified trajectory for specification evidence, refining vague bug reports into structured specifications that guide automated program-repair loops. | The resource is directly reusable as a starting artifact. TrajSpec runs a trajectory-collection agent over the pre-fix repository and mines the unverified trajectory for specification evidence, refining vague bug reports into structured specifications that guide automated program-repair loops. | Use Bug Report Specification Refinement with Trajectory Guidance for Automated Program Repair to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.07882; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,013 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1013 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.07882 | [2607.07882] Bug Report Specification Refinement with Trajectory Guidance for Automated Program Repair | Bug reports serve as task specifications for repository-level automated program repair (APR) agents, but they often describe only the observed failure and omit repair-relevant information such as the failure-inducing behavior, behavioral requirement, and implementation scope. As a result, a repair agent may inspect irr... | S M Farah Al Fahim; Md Nakhla Rafi; Md Ahasanuzzaman; Zeyang Ma; Dong Jae Kim; Shaowei Wang; Tse-Hsun; Chen | 2026-07-08 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.07882 | 2026-07-29T08:06:54 | |||||||||
ale-0369 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Failure as a Process: An Anatomy of CLI Coding Agent Trajectories | https://arxiv.org/abs/2607.09510 | external | arxiv.org | Empirical anatomy of 3,843 CLI coding-agent trajectories across seven models and three scaffolds (OpenHands, MiniSWE, Terminus2), with 1,794 fully annotated over 63,000+ manually reviewed steps; models failure as a temporal process of onset, evolution, and recovery and finds failures dominated by epistemic errors that ... | Empirical anatomy of 3,843 CLI coding-agent trajectories across seven models and three scaffolds (OpenHands, MiniSWE, Terminus2), with 1,794 fully annotated over 63,000+ manually reviewed steps; models failure as a temporal process of onset, evolution, and recovery and finds failures dominated by epistemic errors that ... | Empirical anatomy of 3,843 CLI coding-agent trajectories across seven models and three scaffolds (OpenHands, MiniSWE, Terminus2), with 1,794 fully annotated over 63,000+ manually reviewed steps; models failure as a temporal process of onset, evolution, and recovery and finds failures dominated by epistemic errors that ... | Evaluation data is used as the feedback signal for improving loop behavior. Empirical anatomy of 3,843 CLI coding-agent trajectories across seven models and three scaffolds (OpenHands, MiniSWE, Terminus2), with 1,794 fully annotated over 63,000+ manually reviewed steps; models failure as a temporal process of onset, ev... | Use Failure as a Process: An Anatomy of CLI Coding Agent Trajectories to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.09510; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,014 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1014 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.09510 | [2607.09510] Failure as a Process: An Anatomy of CLI Coding Agent Trajectories | Large language model (LLM) coding agents are increasingly deployed to autonomously perform software engineering tasks in terminal-based environments, making their reliability a growing concern. Existing empirical studies investigate why coding agents fail, yet they largely treat failure as a final outcome rather than a... | Xiangxin Zhao; Han Li; Shuaiting Li; Tianyi Zhao; Earl T. Barr; Federica Sarro; He Ye | 2026-07-10 | 2026 | arXiv | arXiv | 12 pages, 6 figures | cs.SE | arxiv-api | 2607.09510 | 2026-07-29T08:06:54 | ||||||||
ale-0370 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Agentic Proof and Property-Based Testing via Property-Templates | https://arxiv.org/abs/2607.09072 | external | arxiv.org | Dual-track verification-in-the-loop for AI-generated code: shared property templates drive both formal proof in Lean 4 and executable property-based tests for PySpark, raising agentic proof success up to 2.6x and cutting proof hallucinations by 59%. | Dual-track verification-in-the-loop for AI-generated code: shared property templates drive both formal proof in Lean 4 and executable property-based tests for PySpark, raising agentic proof success up to 2.6x and cutting proof hallucinations by 59%. | Dual-track verification-in-the-loop for AI-generated code: shared property templates drive both formal proof in Lean 4 and executable property-based tests for PySpark, raising agentic proof success up to 2.6x and cutting proof hallucinations by 59%. | Verification is promoted from a final check to a loop-control signal. Dual-track verification-in-the-loop for AI-generated code: shared property templates drive both formal proof in Lean 4 and executable property-based tests for PySpark, raising agentic proof success up to 2.6x and cutting proof hallucinations by 59%. | Use Agentic Proof and Property-Based Testing via Property-Templates to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.09072; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,015 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1015 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.09072 | [2607.09072] Agentic Proof and Property-Based Testing via Property-Templates in Data-Intensive Computing | As the cost of code generation becomes cheaper with AI, the new bottleneck in software engineering has shifted to intent specification and validation. Overcoming this durability crisis of AI-driven coding requires more than traditional fuzzing: each candidate property must be proven correct over a model and shown to ho... | Seongmin Lee; Yaoxuan Wu; Miryung Kim | 2026-07-10 | 2026 | arXiv | arXiv | 12 pages, 7 figures, 4 tables; supplementary material included as ancillary file | cs.SE | arxiv-api | 2607.09072 | 2026-07-29T08:06:54 | ||||||||
ale-0371 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP | https://arxiv.org/abs/2607.11098 | external | arxiv.org | Workbench that reproduces an agent failure, intervenes at the point it went wrong, and tests mitigations, turning one-off agent bugs into a repeatable diagnose-and-fix loop over MCP tool use. | Workbench that reproduces an agent failure, intervenes at the point it went wrong, and tests mitigations, turning one-off agent bugs into a repeatable diagnose-and-fix loop over MCP tool use. | Workbench that reproduces an agent failure, intervenes at the point it went wrong, and tests mitigations, turning one-off agent bugs into a repeatable diagnose-and-fix loop over MCP tool use. | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Workbench that reproduces an agent failure, intervenes at the point it went wrong, and tests mitigations, turning one-off agent bugs into a repeatable diagnose-and-fix loop over MCP tool use. | Use AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.11098; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,016 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1016 | 2026-07-15 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.11098 | [2607.11098] AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP | Tool-using LLM agents are mostly evaluated assuming all tools work. When a tool times out, returns a week-stale value, or has its description poisoned in deployment, the developer needs a controlled way to reproduce the failure, test a fix, and confirm the fix worked before deployment. We present AgentCheck, an open-so... | Aritra Mazumder; Nusrat jahan Lia | 2026-07-13 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.11098 | 2026-07-29T08:06:54 | ||||||||
ale-0372 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Latent Programming Horizons in Coding Agents | https://arxiv.org/abs/2607.05188 | external | arxiv.org | Shows a coding agent's hidden states linearly encode program properties like correctness and test outcomes and predict future edits up to 25 steps ahead, a latent signal that could gate or steer verification loops before edits materialize. | Shows a coding agent's hidden states linearly encode program properties like correctness and test outcomes and predict future edits up to 25 steps ahead, a latent signal that could gate or steer verification loops before edits materialize. | Shows a coding agent's hidden states linearly encode program properties like correctness and test outcomes and predict future edits up to 25 steps ahead, a latent signal that could gate or steer verification loops before edits materialize. | Verification is promoted from a final check to a loop-control signal. Shows a coding agent's hidden states linearly encode program properties like correctness and test outcomes and predict future edits up to 25 steps ahead, a latent signal that could gate or steer verification loops before edits materialize. | Use Latent Programming Horizons in Coding Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.05188; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,017 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1017 | 2026-07-15 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.05188 | [2607.05188] Latent Programming Horizons in Coding Agents | A coding agent solving a software-engineering task spends dozens of steps reasoning, editing code, and running tests, yet little is known about what the underlying language model internally represents about the program it is working on. We show that the residual streams of language models under coding agents linearly e... | André Silva; Han Tu; Martin Monperrus | 2026-07-06 | 2026 | arXiv | arXiv | cs.LG | arxiv-api | 2607.05188 | 2026-07-29T08:06:54 | ||||||||
ale-0373 | Verification And Feedback Gates | verification-and-feedback-gates | Docs | 📚 | Why evaluate agents | https://adk.dev/evaluate/ | external | adk.dev | Official ADK guide to evaluating final responses and trajectories, defining test cases, selecting criteria, and running repeatable agent evaluations locally or in CI. | Official ADK guide to evaluating final responses and trajectories, defining test cases, selecting criteria, and running repeatable agent evaluations locally or in CI. | Official ADK guide to evaluating final responses and trajectories, defining test cases, selecting criteria, and running repeatable agent evaluations locally or in CI. | Primary-source operational guidance rather than commentary. Official ADK guide to evaluating final responses and trajectories, defining test cases, selecting criteria, and running repeatable agent evaluations locally or in CI. | Use Why evaluate agents to measure progress and gate completion with repeatable evidence. | Primary official documentation from adk.dev; use it for current product or standard behavior. | high | README.md | 1,018 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1018 | 2026-07-17 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | official-documentation | A | ok | https://adk.dev/evaluate/ | Why evaluate agents - Agent Development Kit (ADK) Agent Development Kit (ADK) | Build powerful multi-agent systems with Agent Development Kit (ADK) | Google Agent Development Kit | Google Agent Development Kit | Google | primary-page | 2026-07-29T08:06:54 | ||||||||||||
ale-0374 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Structured Feedback Improves Repair in an LLM Agent Loop | https://arxiv.org/abs/2607.14167 | external | arxiv.org | In 50 paired TextWorld tasks under a four-call budget, feedback containing the failure location, observed value, and admissible alternatives raises repair success from 14/50 to 36/50 for one model and 8/50 to 29/50 for another; ablations identify alternatives, not JSON syntax, as the main driver. | In 50 paired TextWorld tasks under a four-call budget, feedback containing the failure location, observed value, and admissible alternatives raises repair success from 14/50 to 36/50 for one model and 8/50 to 29/50 for another; ablations identify alternatives, not JSON syntax, as the main driver. | In 50 paired TextWorld tasks under a four-call budget, feedback containing the failure location, observed value, and admissible alternatives raises repair success from 14/50 to 36/50 for one model and 8/50 to 29/50 for another; ablations identify alternatives, not JSON syntax, as the main driver. | The contribution is machine-readable and validation-friendly. In 50 paired TextWorld tasks under a four-call budget, feedback containing the failure location, observed value, and admissible alternatives raises repair success from 14/50 to 36/50 for one model and 8/50 to 29/50 for another; ablations identify alternative... | Use Structured Feedback Improves Repair in an LLM Agent Loop to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.14167; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,019 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1019 | 2026-07-17 | Verify | verify | Gate progress with tests, evals, and evidence. | budget | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.14167 | [2607.14167] Structured Feedback Improves Repair in an LLM Agent Loop | LLM agents often retry after external validation rejects a candidate, but the interface between validation and the next model call remains underspecified. We introduce VeriHarness, a code-controlled agent loop in which models generate candidates while external validators control acceptance, budgets, and traces. We use ... | Jaideep Ray; Ankit Goyal | 2026-07-15 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.14167 | 2026-07-29T08:06:54 | ||||||||
ale-0375 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Copy-on-Write Scoring: Application-Specific Agent Evaluations | https://arxiv.org/abs/2607.14336 | external | arxiv.org | Uses PostgreSQL copy-on-write isolation to let an agent modify a realistic application state while a scorer evaluates the resulting operations safely; the Plane case study also exposes tool-surface defects that simpler task checks miss. | Uses PostgreSQL copy-on-write isolation to let an agent modify a realistic application state while a scorer evaluates the resulting operations safely; the Plane case study also exposes tool-surface defects that simpler task checks miss. | Uses PostgreSQL copy-on-write isolation to let an agent modify a realistic application state while a scorer evaluates the resulting operations safely; the Plane case study also exposes tool-surface defects that simpler task checks miss. | State persistence is explicit enough for repeated runs and handoff. Uses PostgreSQL copy-on-write isolation to let an agent modify a realistic application state while a scorer evaluates the resulting operations safely; the Plane case study also exposes tool-surface defects that simpler task checks miss. | Use Copy-on-Write Scoring: Application-Specific Agent Evaluations to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.14336; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,020 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1020 | 2026-07-17 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;state | researcher;evaluator | harness | enabling | research-paper | A | ok | https://arxiv.org/abs/2607.14336 | [2607.14336] Copy-on-Write Scoring: Application-Specific Agent Evaluations | Trustworthy deployment of LLM-based agents in software systems requires evaluating how they perform on application-specific workflows, with enough granularity to localize where they succeed and fail. Yet existing agent evaluation mechanisms are limited: benchmarks have low construct validity for application-specific wo... | Joanna Roy; Sven Hoelzel | 2026 | 2026 | ICML Workshop on Agents in the Wild: Safety Security and Beyond | International Conference on Machine Learning | Accepted at ICML Workshop on Agents in the Wild: Safety Security and Beyond; the linked arXiv record is the available paper version. | cs.SE | Current arXiv acceptance note and official workshop page | 2607.14336 | 2026-07-29T08:06:54 | |||||||
ale-0376 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | The Prover Is the Judge: Verified Security Software from AI Coding Agents in Ada/SPARK | https://arxiv.org/abs/2607.14340 | external | arxiv.org | Places formal proof obligations inside a coding-agent repair loop and reports 49,280 discharged obligations with 20-40x less supervision; the paper also states that proofs must be paired with known-answer tests, interoperability checks, and human specification review. | Places formal proof obligations inside a coding-agent repair loop and reports 49,280 discharged obligations with 20-40x less supervision; the paper also states that proofs must be paired with known-answer tests, interoperability checks, and human specification review. | Places formal proof obligations inside a coding-agent repair loop and reports 49,280 discharged obligations with 20-40x less supervision; the paper also states that proofs must be paired with known-answer tests, interoperability checks, and human specification review. | Verification is promoted from a final check to a loop-control signal. Places formal proof obligations inside a coding-agent repair loop and reports 49,280 discharged obligations with 20-40x less supervision; the paper also states that proofs must be paired with known-answer tests, interoperability checks, and human spe... | Use The Prover Is the Judge: Verified Security Software from AI Coding Agents in Ada/SPARK to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.14340; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,021 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1021 | 2026-07-17 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.14340 | [2607.14340] The Prover Is the Judge: Verified Security Software from AI Coding Agents in Ada/SPARK | AI coding agents produce code faster than humans can review it. In our approach, the prover is the judge of whether the code is correct. Under a verifier-driven loop, AI agents wrote and verified bare-metal security software in Ada/SPARK spanning classical and post-quantum cryptography, TLS 1.3, IKEv2, X.509, and a Mat... | Tobias Philipp | 2026-07-15 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.14340 | 2026-07-29T08:06:54 | ||||||||
ale-0377 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Verified LLM-Driven Synthesis for Concept Design | https://arxiv.org/abs/2607.15718 | external | arxiv.org | Combines formal concept-and-reaction semantics with a counterexample-guided LLM synthesis loop whose candidates must satisfy machine-checked invariants; experiments on three applications show scenarios steer the loop more consistently than natural-language prompts, while sparse scenarios overfit and nondeterministic om... | Combines formal concept-and-reaction semantics with a counterexample-guided LLM synthesis loop whose candidates must satisfy machine-checked invariants; experiments on three applications show scenarios steer the loop more consistently than natural-language prompts, while sparse scenarios overfit and nondeterministic om... | Combines formal concept-and-reaction semantics with a counterexample-guided LLM synthesis loop whose candidates must satisfy machine-checked invariants; experiments on three applications show scenarios steer the loop more consistently than natural-language prompts, while sparse scenarios overfit and nondeterministic om... | Verification is promoted from a final check to a loop-control signal. Combines formal concept-and-reaction semantics with a counterexample-guided LLM synthesis loop whose candidates must satisfy machine-checked invariants; experiments on three applications show scenarios steer the loop more consistently than natural-la... | Use Verified LLM-Driven Synthesis for Concept Design to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.15718; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,022 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1022 | 2026-07-20 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.15718 | [2607.15718] Verified LLM-Driven Synthesis for Concept Design | Concept Design structures software systems around concepts: user-facing, self-contained units of functionality with a focused purpose. Concepts are composed into applications using synchronization rules called reactions, which specify how actions in one concept trigger actions in others. This paper first gives a formal... | Alcino Cunha | 2026-07-17 | 2026 | arXiv | arXiv | 27 pages, 3 figures | cs.SE | arxiv-api | 2607.15718 | 2026-07-29T08:06:54 | |||||||
ale-0378 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | AEGIS: Assay-Aware Protocol Validation and Runtime Monitoring for Open-Source Liquid Handling Robots | https://arxiv.org/abs/2607.15620 | external | arxiv.org | Pairs preflight LLM checks against machine-readable assay rules with runtime visual monitoring of physical execution; reports adjusted F1 0.97 for protocol validation and average precision 0.89 for trajectory monitoring, while explicitly surfacing the small-pipette and transparent-liquid limits. | Pairs preflight LLM checks against machine-readable assay rules with runtime visual monitoring of physical execution; reports adjusted F1 0.97 for protocol validation and average precision 0.89 for trajectory monitoring, while explicitly surfacing the small-pipette and transparent-liquid limits. | Pairs preflight LLM checks against machine-readable assay rules with runtime visual monitoring of physical execution; reports adjusted F1 0.97 for protocol validation and average precision 0.89 for trajectory monitoring, while explicitly surfacing the small-pipette and transparent-liquid limits. | The contribution is machine-readable and validation-friendly. Pairs preflight LLM checks against machine-readable assay rules with runtime visual monitoring of physical execution; reports adjusted F1 0.97 for protocol validation and average precision 0.89 for trajectory monitoring, while explicitly surfacing the small-... | Use AEGIS: Assay-Aware Protocol Validation and Runtime Monitoring for Open-Source Liquid Handling Robots to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.15620; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,023 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1023 | 2026-07-20 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.15620 | [2607.15620] AEGIS: Assay-Aware Protocol Validation and Runtime Monitoring for Open-Source Liquid Handling Robots | Self-driving laboratories increasingly rely on low-cost liquid handlers such as the Opentrons OT-2, which ship without the pressure-based aspiration monitoring of Hamilton or Tecan systems and are typically run open-loop. Two failure modes go undetected: protocols that are syntactically valid but violate assay-specific... | Priyanka V. Setty; Arvind Ramanathan; Ian Foster; Rick Stevens | 2026-07-17 | 2026 | arXiv | arXiv | cs.RO | arxiv-api | 2607.15620 | 2026-07-29T08:06:54 | ||||||||
ale-0379 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification | https://arxiv.org/abs/2603.02798 | external | arxiv.org | Compiles expert guidelines into step-wise trajectory checks, calibrates accumulated evidence into correctness probabilities, and triggers additional verification when uncertainty remains high. | Compiles expert guidelines into step-wise trajectory checks, calibrates accumulated evidence into correctness probabilities, and triggers additional verification when uncertainty remains high. | Compiles expert guidelines into step-wise trajectory checks, calibrates accumulated evidence into correctness probabilities, and triggers additional verification when uncertainty remains high. | Verification is promoted from a final check to a loop-control signal. Compiles expert guidelines into step-wise trajectory checks, calibrates accumulated evidence into correctness probabilities, and triggers additional verification when uncertainty remains high. | Use GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification to measure progress and gate completion with repeatable evidence. | Research source arXiv:2603.02798; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,027 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1027 | 2026-07-18 | Verify | verify | Gate progress with tests, evals, and evidence. | trigger;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2603.02798 | [2603.02798] Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification | As LLM-powered agents have been used for high-stakes decision-making, such as clinical diagnosis, it becomes critical to develop reliable verification of their decisions to facilitate trustworthy deployment. Yet, existing verifiers usually underperform owing to a lack of domain knowledge and limited calibration. To add... | Yichi Zhang; Nabeel Seedat; Yinpeng Dong; Peng Cui; Jun Zhu; Mihaela van de Schaar | 2026-03-03 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2603.02798 | 2026-07-29T08:06:54 | ||||||||
ale-0380 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Zombie Agents: Detecting Semantic Livelock in Long-Horizon Autonomous Software | https://doi.org/10.1145/3805760.3814895 | external | doi.org | Defines semantic livelock as continued agent activity without progress and proposes an independent embedding-based convergence monitor that detected the pattern in 25% of the analyzed long-duration SWE-agent failures. | Defines semantic livelock as continued agent activity without progress and proposes an independent embedding-based convergence monitor that detected the pattern in 25% of the analyzed long-duration SWE-agent failures. | Defines semantic livelock as continued agent activity without progress and proposes an independent embedding-based convergence monitor that detected the pattern in 25% of the analyzed long-duration SWE-agent failures. | The work targets tasks that exceed a single context window or prompt session. Defines semantic livelock as continued agent activity without progress and proposes an independent embedding-based convergence monitor that detected the pattern in 25% of the analyzed long-duration SWE-agent failures. | Use Zombie Agents: Detecting Semantic Livelock in Long-Horizon Autonomous Software to measure progress and gate completion with repeatable evidence. | Research source; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,028 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1028 | 2026-07-18 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-paper | A | restricted | https://doi.org/10.1145/3805760.3814895 | Simarjot Khanna | 2026-07 | 2026 | Proceedings of the 3rd ACM International Conference on AI-Powered Software (AIware '26) | Association for Computing Machinery | 10.1145/3805760.3814895 | Published at AIware 2026; metadata verified from the author-supplied camera-ready paper because the DOI landing page restricted automated access. | ACM DOI and camera-ready paper | 2026-07-29T08:06:54 | ||||||||||
ale-0381 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents | https://arxiv.org/abs/2607.12790 | external | arxiv.org | Every self-improvement loop rests on an evaluation metric that can itself drift or be gamed; the Double Ratchet framework co-evolves metrics alongside agent skills with anchor discipline and independent audits, retaining 88-110% of ground-truth lift across code generation, SQL, and report-writing loops. | Every self-improvement loop rests on an evaluation metric that can itself drift or be gamed; the Double Ratchet framework co-evolves metrics alongside agent skills with anchor discipline and independent audits, retaining 88-110% of ground-truth lift across code generation, SQL, and report-writing loops. | Every self-improvement loop rests on an evaluation metric that can itself drift or be gamed; the Double Ratchet framework co-evolves metrics alongside agent skills with anchor discipline and independent audits, retaining 88-110% of ground-truth lift across code generation, SQL, and report-writing loops. | Evaluation data is used as the feedback signal for improving loop behavior. Every self-improvement loop rests on an evaluation metric that can itself drift or be gamed; the Double Ratchet framework co-evolves metrics alongside agent skills with anchor discipline and independent audits, retaining 88-110% of ground-truth... | Use Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.12790; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,029 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1029 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.12790 | [2607.12790] Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents | Self-evolving agent systems improve by creating, revising, and retiring their own skills, but every such loop rests on a hidden assumption: a reliable evaluation metric already exists. In many real applications it does not. We make three claims. First, metrics can be \emph{evolved}: our metric loop searches composition... | Xing Zhang; Guanghui Wang; Yanwei Cui; Ziyuan Li; Wei Qiu; Bing Zhu; Peiyang He | 2026-07-14 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.12790 | 2026-07-29T08:06:54 | ||||||||
ale-0382 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | AgentLTL: A Trace-Verification Framework for Measuring, Enforcing, and Training Procedural Compliance in Tool-Using LLM Agents | https://arxiv.org/abs/2607.02599 | external | arxiv.org | First-Order Linear Temporal Logic specification language for verifying that tool-using agents follow procedural rules across the whole trace, not just reach correct answers; supports real-time in-loop tool-call gating and dense-reward training, with compliance gains that generalize to unseen procedural variations. | First-Order Linear Temporal Logic specification language for verifying that tool-using agents follow procedural rules across the whole trace, not just reach correct answers; supports real-time in-loop tool-call gating and dense-reward training, with compliance gains that generalize to unseen procedural variations. | First-Order Linear Temporal Logic specification language for verifying that tool-using agents follow procedural rules across the whole trace, not just reach correct answers; supports real-time in-loop tool-call gating and dense-reward training, with compliance gains that generalize to unseen procedural variations. | Verification is promoted from a final check to a loop-control signal. First-Order Linear Temporal Logic specification language for verifying that tool-using agents follow procedural rules across the whole trace, not just reach correct answers; supports real-time in-loop tool-call gating and dense-reward training, with ... | Use AgentLTL: A Trace-Verification Framework for Measuring, Enforcing, and Training Procedural Compliance in Tool-Using LLM Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.02599; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,030 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1030 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.02599 | [2607.02599] AgentLTL: A Trace-Verification Framework for Measuring, Enforcing, and Training Procedural Compliance in Tool-Using LLM Agents | Tool-using LLM agents are usually evaluated by final-answer correctness or LLM judges. Neither captures how an answer was produced. In safety-critical settings, the procedure itself is part of correctness. In this paper, we introduce AgentLTL, a language derived from First-Order Linear Temporal Logic (FO-LTL) that expr... | Laïla Elkoussy; Julien Perez | 2026-07-01 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.02599 | 2026-07-29T08:06:54 | ||||||||
ale-0383 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade | https://arxiv.org/abs/2607.06503 | external | arxiv.org | Trains lightweight linear probes on hidden states from an episode's earliest interactions to predict eventual failure, then aborts doomed runs through a calibrated cascade with user-specified recall guarantees, cutting generated tokens by roughly 54-60% at a 90% recall target on TextCraft and WebShop across three model... | Trains lightweight linear probes on hidden states from an episode's earliest interactions to predict eventual failure, then aborts doomed runs through a calibrated cascade with user-specified recall guarantees, cutting generated tokens by roughly 54-60% at a 90% recall target on TextCraft and WebShop across three model... | Trains lightweight linear probes on hidden states from an episode's earliest interactions to predict eventual failure, then aborts doomed runs through a calibrated cascade with user-specified recall guarantees, cutting generated tokens by roughly 54-60% at a 90% recall target on TextCraft and WebShop across three model... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Trains lightweight linear probes on hidden states from an episode's earliest interactions to predict eventual failure, then aborts doomed runs through a calibrated cascade with user-specified recall guarantees, cutting generated tokens by ro... | Use Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.06503; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,031 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1031 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | budget;exit | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.06503 | [2607.06503] Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade | Large language model (LLM) agents often waste inference compute by continuing multi-step trajectories that are already doomed to fail. We study early failure prediction and inference-time early stopping for LLM agents using hidden-state probes. Lightweight linear probes on internal activations predict eventual task fai... | Kai Ruan; Zihe Huang; Ziqi Zhou; Qianshan Wei; Jinghao Lin; Xuan Wang; Hao Sun | 2026-07-07 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.06503 | 2026-07-29T08:06:54 | ||||||||
ale-0384 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Where Does Agent Reliability Come From? A Cross-Benchmark Decomposition of Verification Loops, Specialist Models, and Scaffolding in a Production Enterprise Agent | https://arxiv.org/abs/2607.17044 | external | arxiv.org | Ablates a production enterprise agent (Leni) across three benchmarks to decompose reliability gains among verification loops, specialist models, and scaffolding: 7-15pp improvements come mostly from scaffolding, routing, and specialist models, while the verification step's isolated gain is small (+1.5pp) yet rescues ot... | Ablates a production enterprise agent (Leni) across three benchmarks to decompose reliability gains among verification loops, specialist models, and scaffolding: 7-15pp improvements come mostly from scaffolding, routing, and specialist models, while the verification step's isolated gain is small (+1.5pp) yet rescues ot... | Ablates a production enterprise agent (Leni) across three benchmarks to decompose reliability gains among verification loops, specialist models, and scaffolding: 7-15pp improvements come mostly from scaffolding, routing, and specialist models, while the verification step's isolated gain is small (+1.5pp) yet rescues ot... | Verification is promoted from a final check to a loop-control signal. Ablates a production enterprise agent (Leni) across three benchmarks to decompose reliability gains among verification loops, specialist models, and scaffolding: 7-15pp improvements come mostly from scaffolding, routing, and specialist models, while ... | Use Where Does Agent Reliability Come From? A Cross-Benchmark Decomposition of Verification Loops, Specialist Models, and Scaffolding in a Production Enterprise Agent to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.17044; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,032 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1032 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.17044 | [2607.17044] Where Does Agent Reliability Come From? A Cross-Benchmark Decomposition of Verification Loops, Specialist Models, and Scaffolding in a Production Enterprise Agent | Multi-step enterprise agent tasks fail in a characteristic way: single-pass inference has no checkpoint between deciding an answer and committing to it. We study one production system (Leni) whose architecture installs such checkpoints: verification loops (execute, observe, compare, correct) staffed by lightweight task... | Arunabh Dastidar | 2026-07-19 | 2026 | arXiv | arXiv | 19 pages, 5 figures, 5 tables. Evaluations conducted March-April 2026. Run-level evaluation record and audit scripts: https://github.com/arnabdastidar/leni-agent-evals | cs.SE | arxiv-api | 2607.17044 | 2026-07-29T08:06:54 | |||||||
ale-0385 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Test Coverage Analysis of Agentic Pull Requests | https://arxiv.org/abs/2607.18057 | external | arxiv.org | Mines 4,882 agent-generated PRs (five coding agents) from the AIDev dataset and finds agents ship tests in only 49.6% of code-changing PRs, existing tests cover just 27% of changed Python lines, and error-handling code goes unexecuted at up to 86% miss rates, empirical evidence that unattended agent loops under-verify ... | Mines 4,882 agent-generated PRs (five coding agents) from the AIDev dataset and finds agents ship tests in only 49.6% of code-changing PRs, existing tests cover just 27% of changed Python lines, and error-handling code goes unexecuted at up to 86% miss rates, empirical evidence that unattended agent loops under-verify ... | Mines 4,882 agent-generated PRs (five coding agents) from the AIDev dataset and finds agents ship tests in only 49.6% of code-changing PRs, existing tests cover just 27% of changed Python lines, and error-handling code goes unexecuted at up to 86% miss rates, empirical evidence that unattended agent loops under-verify ... | Packages the evidence as queryable CSV and JSONL rather than only a rendered page. Mines 4,882 agent-generated PRs (five coding agents) from the AIDev dataset and finds agents ship tests in only 49.6% of code-changing PRs, existing tests cover just 27% of changed Python lines, and error-handling code goes unexecuted at... | Use Test Coverage Analysis of Agentic Pull Requests to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.18057; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,033 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1033 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.18057 | [2607.18057] Test Coverage Analysis of Agentic Pull Requests | AI coding agents increasingly submit complete pull requests (PRs) with minimal human intervention, shifting software development from AI-assisted to autonomous workflows. As these agents become more prevalent, ensuring the code they generate is adequately tested, by existing tests or by tests the agents write, is criti... | Atish Kumar Dipongkor; Talank Baral; Wing Lam; Kevin Moran | 2026-07-20 | 2026 | arXiv | arXiv | 12 pages, to appear 42nd International Conference on Software Maintenance and Evolution | cs.SE | arxiv-api | 2607.18057 | 2026-07-29T08:06:54 | |||||||
ale-0386 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | TRIM: Reducing AI-Generated CodeSlop via Agent Trajectory Minimization | https://arxiv.org/abs/2607.18161 | external | arxiv.org | Attributes verbose "CodeSlop" to the agent loop's own search process, speculative edits, abandoned hypotheses, and temporary changes that survive into the final diff, and removes 17.9-32.9% of redundant code across agentic scaffolds by minimizing the trajectory rather than post-editing the code, with negligible perform... | Attributes verbose "CodeSlop" to the agent loop's own search process, speculative edits, abandoned hypotheses, and temporary changes that survive into the final diff, and removes 17.9-32.9% of redundant code across agentic scaffolds by minimizing the trajectory rather than post-editing the code, with negligible perform... | Attributes verbose "CodeSlop" to the agent loop's own search process, speculative edits, abandoned hypotheses, and temporary changes that survive into the final diff, and removes 17.9-32.9% of redundant code across agentic scaffolds by minimizing the trajectory rather than post-editing the code, with negligible perform... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Attributes verbose "CodeSlop" to the agent loop's own search process, speculative edits, abandoned hypotheses, and temporary changes that survive into the final diff, and removes 17.9-32.9% of redundant code across agentic scaffolds by minim... | Use TRIM: Reducing AI-Generated CodeSlop via Agent Trajectory Minimization to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.18161; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,034 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1034 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.18161 | [2607.18161] TRIM: Reducing AI-Generated CodeSlop via Agent Trajectory Minimization | Coding agents are increasingly used to accelerate code generation in many downstream tasks, such as fixing bugs, building applications, and prototyping. However, despite their value as coding assistants, agent-generated code tends to be larger and more verbose than the corresponding human-written implementation. In thi... | Alex Mathai; Shobini Iyer; Aleksandr Nogikh; Petros Maniatis; Franjo Ivancic; Junfeng Yang; Baishakhi Ray | 2026-07-20 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.18161 | 2026-07-29T08:06:54 | ||||||||
ale-0387 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Agentic Code Review in the Terminal: A Trajectory-Level Analysis of Behavior, Cost, and Human-Alignment | https://arxiv.org/abs/2607.16740 | external | arxiv.org | Trajectory-level study of terminal-based code-review agents that give feedback before pull-request creation, finding they achieve higher review precision but incur substantial exploration and validation overhead, and that successful reviews are associated with stronger planning and less downstream validation, behavior ... | Trajectory-level study of terminal-based code-review agents that give feedback before pull-request creation, finding they achieve higher review precision but incur substantial exploration and validation overhead, and that successful reviews are associated with stronger planning and less downstream validation, behavior ... | Trajectory-level study of terminal-based code-review agents that give feedback before pull-request creation, finding they achieve higher review precision but incur substantial exploration and validation overhead, and that successful reviews are associated with stronger planning and less downstream validation, behavior ... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Trajectory-level study of terminal-based code-review agents that give feedback before pull-request creation, finding they achieve higher review precision but incur substantial exploration and validation overhead, and that successful reviews ... | Use Agentic Code Review in the Terminal: A Trajectory-Level Analysis of Behavior, Cost, and Human-Alignment to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.16740; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,035 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1035 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | budget;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.16740 | [2607.16740] Agentic Code Review in the Terminal: A Trajectory-Level Analysis of Behavior, Cost, and Human-Alignment | Agentic code review in terminal-based environments enables early feedback during local development before pull request creation. However, existing evaluations remain performance-centric and fail to capture the dynamic behaviors of repository-grounded agentic reviewers. Understanding these behaviors is critical for iden... | Wachiraphan Charoenwet; Kla Tantithamthavorn; Patanamon Thongtanunam; Hong Yi Lin; Minwoo Jeong; Ming Wu | 2026-07-18 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.16740 | 2026-07-29T08:06:54 | ||||||||
ale-0388 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | brain0 | https://github.com/Brain0-ai/brain0 | external | github.com | Black-box decision graph for AI-written code that records why an agent made each change, making agent-authored diffs inspectable after the fact. | Black-box decision graph for AI-written code that records why an agent made each change, making agent-authored diffs inspectable after the fact. | Black-box decision graph for AI-written code that records why an agent made each change, making agent-authored diffs inspectable after the fact. | Control flow is represented as an inspectable graph rather than an opaque prompt loop. Black-box decision graph for AI-written code that records why an agent made each change, making agent-authored diffs inspectable after the fact. | Use brain0 to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (398 stars; 13 forks; Apache-2.0 license; updated 2026-07-28); popularity is context, not proof of reliability. | medium | README.md | 1,036 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1036 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/Brain0-ai/brain0 | GitHub - Brain0-ai/brain0: The black box for AI-written code. Passive decision graph linking every commit to the agent prompts behind it: drift detection, DLP audit of what agents read, evidence-driven risk, MCP memory for coding agents, signed provenance attestations. One command, offline by default. · GitHub | The black box for AI-written code. Passive decision graph linking every commit to the agent prompts behind it: drift detection, DLP audit of what agents read, evidence-driven risk, MCP memory for coding agents, signed provenance attestations. One command, offline by default. - Brain0-ai/brain0 | 2026-07-02 | 2026 | Brain0-ai/brain0 | GitHub | github-api | Brain0-ai/brain0 | 398 | 13 | Apache-2.0 | 2026-07-02T12:02:08Z | 2026-07-28T12:36:35Z | 2026-07-29T08:06:54 | |||||
ale-0389 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Watch Skill | https://github.com/oxbshw/watch-skill | external | github.com | Video understanding and self-verification layer that turns videos, streams, and agent screen recordings into searchable evidence an agent can check its own work against. | Video understanding and self-verification layer that turns videos, streams, and agent screen recordings into searchable evidence an agent can check its own work against. | Video understanding and self-verification layer that turns videos, streams, and agent screen recordings into searchable evidence an agent can check its own work against. | The agent workflow includes explicit self-checking or gated completion. Video understanding and self-verification layer that turns videos, streams, and agent screen recordings into searchable evidence an agent can check its own work against. | Use Watch Skill to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (238 stars; 36 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 1,037 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1037 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/oxbshw/watch-skill | GitHub - oxbshw/watch-skill: Video understanding and self-verification for AI agents. Turn videos, streams, and agent screen recordings into searchable, timestamped evidence—then use THE LOOP to inspect, fix, and verify the work. MCP, CLI, REST, local-first. · GitHub | Video understanding and self-verification for AI agents. Turn videos, streams, and agent screen recordings into searchable, timestamped evidence—then use THE LOOP to inspect, fix, and verify the work. MCP, CLI, REST, local-first. - oxbshw/watch-skill | 2026-07-05 | 2026 | oxbshw/watch-skill | GitHub | github-api | oxbshw/watch-skill | 238 | 36 | MIT | 2026-07-05T23:25:30Z | 2026-07-29T03:21:55Z | 2026-07-29T08:06:54 | |||||
ale-0390 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Auto | https://github.com/RightNow-AI/auto | external | github.com | Compiles recorded LLM-agent behavior into verified, capability-confined WebAssembly binaries, proving which parts of a loop are secretly symbolic and distilling them into deterministic code. | Compiles recorded LLM-agent behavior into verified, capability-confined WebAssembly binaries, proving which parts of a loop are secretly symbolic and distilling them into deterministic code. | Compiles recorded LLM-agent behavior into verified, capability-confined WebAssembly binaries, proving which parts of a loop are secretly symbolic and distilling them into deterministic code. | Verification is promoted from a final check to a loop-control signal. Compiles recorded LLM-agent behavior into verified, capability-confined WebAssembly binaries, proving which parts of a loop are secretly symbolic and distilling them into deterministic code. | Use Auto to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (116 stars; 10 forks; Apache-2.0 license; updated 2026-07-22); popularity is context, not proof of reliability. | medium | README.md | 1,038 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1038 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/RightNow-AI/auto | GitHub - RightNow-AI/auto: the agi compiler: records llm agent behavior, proves what repeats, and compiles it into verified, sandboxed wasm binaries that run for microdollars. nothing figured out twice, paper: https://arxiv.org/abs/2607.04542 · GitHub | the agi compiler: records llm agent behavior, proves what repeats, and compiles it into verified, sandboxed wasm binaries that run for microdollars. nothing figured out twice, paper: https://arxiv.org/abs/2607.04542 - RightNow-AI/auto | 2026-07-05 | 2026 | RightNow-AI/auto | GitHub | github-api | RightNow-AI/auto | 116 | 10 | Apache-2.0 | 2026-07-05T15:20:57Z | 2026-07-22T12:44:22Z | 2026-07-29T08:06:54 | |||||
ale-0391 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Critic Experience Bank: Self-Evolving Step-Level Confidence Estimation for LLM Agents | https://arxiv.org/abs/2607.12397 | external | arxiv.org | Training-free verification-in-the-loop: a hindsight reviewer labels which steps of completed trajectories were genuinely productive, building an experience bank that calibrates the critic's step-level confidence on future runs, improving when a loop should trust, retry, or escalate an action. | Training-free verification-in-the-loop: a hindsight reviewer labels which steps of completed trajectories were genuinely productive, building an experience bank that calibrates the critic's step-level confidence on future runs, improving when a loop should trust, retry, or escalate an action. | Training-free verification-in-the-loop: a hindsight reviewer labels which steps of completed trajectories were genuinely productive, building an experience bank that calibrates the critic's step-level confidence on future runs, improving when a loop should trust, retry, or escalate an action. | Verification is promoted from a final check to a loop-control signal. Training-free verification-in-the-loop: a hindsight reviewer labels which steps of completed trajectories were genuinely productive, building an experience bank that calibrates the critic's step-level confidence on future runs, improving when a loop ... | Use Critic Experience Bank: Self-Evolving Step-Level Confidence Estimation for LLM Agents to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.12397; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,039 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1039 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;budget;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.12397 | [2607.12397] Critic Experience Bank: Self-Evolving Step-Level Confidence Estimation for LLM Agents | LLM agents act in external environments where each action changes the state that later decisions condition on, and where a single wrong step can waste interaction budget or trigger irreversible side effects long before the final failure is observed. Reliable deployment therefore requires \emph{step-level confidence est... | Yaopei Zeng; Congchao Wang; JianHang Chen; Nan Wang; Yurui Chang; Lu Lin | 2026-07-14 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.12397 | 2026-07-29T08:06:54 | ||||||||
ale-0392 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Tracing Agentic Failure from the Flow of Success | https://arxiv.org/abs/2607.12747 | external | arxiv.org | Failure attribution for agent loops that learns only from successful trajectories: OAT uses neural controlled differential equations to spot the error step in failed runs, 200-5000x faster than prompting-based attribution with ~20% better F1, needing just 100 success traces, makes in-loop failure localization cheap eno... | Failure attribution for agent loops that learns only from successful trajectories: OAT uses neural controlled differential equations to spot the error step in failed runs, 200-5000x faster than prompting-based attribution with ~20% better F1, needing just 100 success traces, makes in-loop failure localization cheap eno... | Failure attribution for agent loops that learns only from successful trajectories: OAT uses neural controlled differential equations to spot the error step in failed runs, 200-5000x faster than prompting-based attribution with ~20% better F1, needing just 100 success traces, makes in-loop failure localization cheap eno... | Treats feedback, telemetry, and deterministic artifacts as loop-control gates. Failure attribution for agent loops that learns only from successful trajectories: OAT uses neural controlled differential equations to spot the error step in failed runs, 200-5000x faster than prompting-based attribution with ~20% better F1... | Use Tracing Agentic Failure from the Flow of Success to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.12747; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,040 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1040 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.12747 | [2607.12747] Tracing Agentic Failure from the Flow of Success | Failure attribution for LLM-based agentic systems, i.e., identifying which steps in a failure trajectory caused the task to fail, is critical for debugging and improving these systems. Existing approaches either rely on prompting-based pipelines, which are computationally expensive, or require post-training on failure ... | Samuel Yeh; Yiwen Zhu; Shaleen Deep; Sharon Li | 2026-07-14 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.12747 | 2026-07-29T08:06:54 | ||||||||
ale-0393 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Beyond Test Presence: Assessing the Quality and Robustness of Agent-Generated Tests | https://arxiv.org/abs/2607.12068 | external | arxiv.org | Study of 204,673 test artifacts comparing agent-generated and human tests: agents win on edge-case and boundary coverage but produce flakier tests (file I/O, non-determinism), a caution that the verification step of a loop can itself become the unreliable component if agent-written checks go unaudited. | Study of 204,673 test artifacts comparing agent-generated and human tests: agents win on edge-case and boundary coverage but produce flakier tests (file I/O, non-determinism), a caution that the verification step of a loop can itself become the unreliable component if agent-written checks go unaudited. | Study of 204,673 test artifacts comparing agent-generated and human tests: agents win on edge-case and boundary coverage but produce flakier tests (file I/O, non-determinism), a caution that the verification step of a loop can itself become the unreliable component if agent-written checks go unaudited. | Verification is promoted from a final check to a loop-control signal. Study of 204,673 test artifacts comparing agent-generated and human tests: agents win on edge-case and boundary coverage but produce flakier tests (file I/O, non-determinism), a caution that the verification step of a loop can itself become the unrel... | Use Beyond Test Presence: Assessing the Quality and Robustness of Agent-Generated Tests to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.12068; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,041 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1041 | 2026-07-22 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.12068 | [2607.12068] Beyond Test Presence: Assessing the Quality and Robustness of Agent-Generated Tests in Open-Source Projects | The integration of AI-powered coding agents into Continuous Integration/Continuous Delivery (CI/CD) pipelines has fundamentally altered how software verification is conducted. While these agents successfully automate the test generation, current evaluation benchmarks (e.g., SWE-bench) largely focus on pass-rates rather... | Preet Jhanglani; Zeel Kaushal Desai; Vidhi Kansara; Eman Abdullah AlOmar | 2026-07-13 | 2026 | arXiv | arXiv | cs.SE | arxiv-api | 2607.12068 | 2026-07-29T08:06:54 | ||||||||
ale-0394 | Verification And Feedback Gates | verification-and-feedback-gates | Blog | 📝 | What a Verification Loop Adds to a Coding Agent: A First Look | https://ironbee.medium.com/what-a-verification-loop-adds-to-a-coding-agent-a-first-look-5049017e636e | external | ironbee.medium.com | First-hand write-up measuring what wrapping a coding agent in an explicit verification loop changes in practice, comparing gated and ungated runs on the same tasks. | First-hand write-up measuring what wrapping a coding agent in an explicit verification loop changes in practice, comparing gated and ungated runs on the same tasks. | First-hand write-up measuring what wrapping a coding agent in an explicit verification loop changes in practice, comparing gated and ungated runs on the same tasks. | Verification is promoted from a final check to a loop-control signal. First-hand write-up measuring what wrapping a coding agent in an explicit verification loop changes in practice, comparing gated and ungated runs on the same tasks. | Use What a Verification Loop Adds to a Coding Agent: A First Look to measure progress and gate completion with repeatable evidence. | Contextual source from ironbee.medium.com; useful for practice signals or boundary conditions, not independent validation. | contextual | README.md | 1,042 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1042 | 2026-07-23 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | evaluator | harness | enabling | practitioner-analysis | B | ok | https://ironbee.medium.com/what-a-verification-loop-adds-to-a-coding-agent-a-first-look-5049017e636e | Medium What a Verification Loop Adds to a Coding Agent: A First Look | by IronBee | Jul, 2026 | Medium | What a Verification Loop Adds to a Coding Agent: A First Look This is the opening post in an ongoing series. We start with one model pair on one project, and the analysis will continue across more … | IronBee | 2026-07-07 | 2026 | Medium | html-meta | 2026-07-29T08:06:54 | |||||||||||
ale-0395 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Best-of-Evidence: Best-of-N Selection under Partial Verification | https://arxiv.org/abs/2607.20950 | external | arxiv.org | Inference-time selection framework replacing whole-response Best-of-N scoring with budgeted claim-level partial verification, using a signed candidate-factor graph so one checkable finding can support part of the candidate pool while contradicting the rest, with proven logarithmic-vs-linear query savings from shared ev... | Inference-time selection framework replacing whole-response Best-of-N scoring with budgeted claim-level partial verification, using a signed candidate-factor graph so one checkable finding can support part of the candidate pool while contradicting the rest, with proven logarithmic-vs-linear query savings from shared ev... | Inference-time selection framework replacing whole-response Best-of-N scoring with budgeted claim-level partial verification, using a signed candidate-factor graph so one checkable finding can support part of the candidate pool while contradicting the rest, with proven logarithmic-vs-linear query savings from shared ev... | Control flow is represented as an inspectable graph rather than an opaque prompt loop. Inference-time selection framework replacing whole-response Best-of-N scoring with budgeted claim-level partial verification, using a signed candidate-factor graph so one checkable finding can support part of the candidate pool while... | Use Best-of-Evidence: Best-of-N Selection under Partial Verification to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.20950; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,043 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1043 | 2026-07-24 | Verify | verify | Gate progress with tests, evals, and evidence. | verification | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.20950 | [2607.20950] Best-of-Evidence: Best-of-N Selection under Partial Verification | BoN improves model outputs by sampling several candidates and selecting one with a proxy score, but it assumes that complete candidates can be evaluated reliably. Many vision-language tasks instead provide only partial verification: a finding, span, value, region, or relation may be checkable even when no dependable wh... | Cenwei Zhang; Teng Fang; Yuxia Wang; Derek Li; Bryan Dai; Lei You | 2026-07-23 | 2026 | arXiv | arXiv | 3 figures, 28 pages | cs.LG | arxiv-api | 2607.20950 | 2026-07-29T08:06:54 | |||||||
ale-0396 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Toward Continuous Assurance for the Democratization of AI Agent Creation in Industry | https://arxiv.org/abs/2607.21495 | external | arxiv.org | Position paper (Levy & Berger, Jul 2026) on the silent-degradation gap of low-code/no-code agents whose dependencies, models, tools, retrieval sources, permissions, schedules, drift after deployment; proposes continuous assurance via dependency mapping, readiness contracts, scheduled checks, and lifecycle governance, w... | Position paper (Levy & Berger, Jul 2026) on the silent-degradation gap of low-code/no-code agents whose dependencies, models, tools, retrieval sources, permissions, schedules, drift after deployment; proposes continuous assurance via dependency mapping, readiness contracts, scheduled checks, and lifecycle governance, w... | Position paper (Levy & Berger, Jul 2026) on the silent-degradation gap of low-code/no-code agents whose dependencies, models, tools, retrieval sources, permissions, schedules, drift after deployment; proposes continuous assurance via dependency mapping, readiness contracts, scheduled checks, and lifecycle governance, w... | The trigger or cadence is explicit, making the workflow recurring rather than one-off. Position paper (Levy & Berger, Jul 2026) on the silent-degradation gap of low-code/no-code agents whose dependencies, models, tools, retrieval sources, permissions, schedules, drift after deployment; proposes continuous assurance via... | Use Toward Continuous Assurance for the Democratization of AI Agent Creation in Industry to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.21495; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,044 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1044 | 2026-07-24 | Verify | verify | Gate progress with tests, evals, and evidence. | trigger;workspace;context | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.21495 | [2607.21495] Toward Continuous Assurance for the Democratization of AI Agent Creation in Industry | AI agents are increasingly created inside organizations by non-engineering users through low-code, no-code, and conversational development environments. This democratization enables rapid local innovation, but it also creates a reliability gap: agents that appear to users as simple productivity artifacts may depend on ... | Natan Levy; Harel Berger | 2026-07-23 | 2026 | arXiv | arXiv | cs.AI | arxiv-api | 2607.21495 | 2026-07-29T08:06:54 | ||||||||
ale-0397 | Verification And Feedback Gates | verification-and-feedback-gates | Docs | 📚 | Catch Security Issues as Claude Writes Code | https://code.claude.com/docs/en/security-guidance | external | code.claude.com | Official docs for Anthropic's security-guidance plugin (July 2026), which wires a verification loop into Claude Code's own lifecycle entirely via hooks: a deterministic per-edit pattern check with no model call, a background fresh-context model review of each turn's Git diff that re-prompts Claude with its findings, an... | Official docs for Anthropic's security-guidance plugin (July 2026), which wires a verification loop into Claude Code's own lifecycle entirely via hooks: a deterministic per-edit pattern check with no model call, a background fresh-context model review of each turn's Git diff that re-prompts Claude with its findings, an... | Official docs for Anthropic's security-guidance plugin (July 2026), which wires a verification loop into Claude Code's own lifecycle entirely via hooks: a deterministic per-edit pattern check with no model call, a background fresh-context model review of each turn's Git diff that re-prompts Claude with its findings, an... | Primary-source operational guidance rather than commentary. Official docs for Anthropic's security-guidance plugin (July 2026), which wires a verification loop into Claude Code's own lifecycle entirely via hooks: a deterministic per-edit pattern check with no model call, a background fresh-context model review of each ... | Use Catch Security Issues as Claude Writes Code to measure progress and gate completion with repeatable evidence. | Primary official documentation from code.claude.com; use it for current product or standard behavior. | high | README.md | 1,045 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1045 | 2026-07-24 | Verify | verify | Gate progress with tests, evals, and evidence. | intake;context;verification | builder;evaluator | harness | enabling | official-documentation | A | ok | https://code.claude.com/docs/en/security-guidance | Catch security issues as Claude writes code - Claude Code Docs | Install the security-guidance plugin to have Claude review its own code changes for vulnerabilities and fix them in the same session. | Claude Code Docs | html-meta | 2026-07-29T08:06:54 | ||||||||||||||
ale-0398 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | pAI-Econ-claude: A Gated Human-in-the-Loop Multi-Agent Architecture for AI-Assisted Economic Theory Development | https://arxiv.org/abs/2607.21268 | external | arxiv.org | Gated human-in-the-loop multi-agent architecture for domains lacking any cheap machine-checkable correctness signal, using economic theory development as the case study. Specialized diagnostic gates, human checkpoints, and a shared workspace of inspectable intermediate records replace full automation; across five tasks... | Gated human-in-the-loop multi-agent architecture for domains lacking any cheap machine-checkable correctness signal, using economic theory development as the case study. Specialized diagnostic gates, human checkpoints, and a shared workspace of inspectable intermediate records replace full automation; across five tasks... | Gated human-in-the-loop multi-agent architecture for domains lacking any cheap machine-checkable correctness signal, using economic theory development as the case study. Specialized diagnostic gates, human checkpoints, and a shared workspace of inspectable intermediate records replace full automation; across five tasks... | Checkpointed state makes long-running agent work recoverable across failures. Gated human-in-the-loop multi-agent architecture for domains lacking any cheap machine-checkable correctness signal, using economic theory development as the case study. Specialized diagnostic gates, human checkpoints, and a shared workspace ... | Use pAI-Econ-claude: A Gated Human-in-the-Loop Multi-Agent Architecture for AI-Assisted Economic Theory Development to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.21268; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,046 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1046 | 2026-07-25 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;delegation;state;escalation | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.21268 | [2607.21268] pAI-Econ-claude: A Gated Human-in-the-Loop Multi-Agent Architecture for AI-Assisted Economic Theory Development | In many social-science research tasks, such as economics, LLM-based agents must produce outputs for which no cheap, task-complete, machine-readable correctness signal exists. This creates a distinctive reliability problem for multi-agent systems: how should generation, critique, coordination, and human judgment be orga... | Chen Zhu; Xiaolu Wang; Weilong Zhang | 2026-07-23 | 2026 | arXiv | arXiv | cs.MA | arxiv-api | 2607.21268 | 2026-07-29T08:06:54 | ||||||||
ale-0399 | Verification And Feedback Gates | verification-and-feedback-gates | Tool | 🧰 | Review Loop | https://github.com/earendil-works/pi-review-loop | external | github.com | First-party extension from earendil-works, the org behind the pi coding agent (77k stars): keeps a native review window open while the agent works, and submitting a review records the current workspace as a session-backed checkpoint so the next pass diffs only the increment ("Since review" vs "vs HEAD" modes), turning ... | First-party extension from earendil-works, the org behind the pi coding agent (77k stars): keeps a native review window open while the agent works, and submitting a review records the current workspace as a session-backed checkpoint so the next pass diffs only the increment ("Since review" vs "vs HEAD" modes), turning ... | First-party extension from earendil-works, the org behind the pi coding agent (77k stars): keeps a native review window open while the agent works, and submitting a review records the current workspace as a session-backed checkpoint so the next pass diffs only the increment ("Since review" vs "vs HEAD" modes), turning ... | Checkpointed state makes long-running agent work recoverable across failures. First-party extension from earendil-works, the org behind the pi coding agent (77k stars): keeps a native review window open while the agent works, and submitting a review records the current workspace as a session-backed checkpoint so the ne... | Use Review Loop to measure progress and gate completion with repeatable evidence. | Inspectable GitHub source (131 stars; 17 forks; MIT license; updated 2026-07-29); popularity is context, not proof of reliability. | medium | README.md | 1,047 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1047 | 2026-07-25 | Verify | verify | Gate progress with tests, evals, and evidence. | workspace;verification;state;escalation | builder;evaluator | harness | enabling | source-implementation | A | ok | https://github.com/earendil-works/pi-review-loop | GitHub - earendil-works/pi-review-loop: A persistent incremental diff review loop for pi · GitHub | A persistent incremental diff review loop for pi. Contribute to earendil-works/pi-review-loop development by creating an account on GitHub. | 2026-07-22 | 2026 | earendil-works/pi-review-loop | GitHub | github-api | earendil-works/pi-review-loop | 131 | 17 | MIT | 2026-07-22T09:44:36Z | 2026-07-29T06:54:29Z | 2026-07-29T08:06:54 | |||||
ale-0400 | Verification And Feedback Gates | verification-and-feedback-gates | Paper | 📄 | Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair | https://arxiv.org/abs/2607.24604 | external | arxiv.org | Directly attacks the core Loop Engineering assumption that more generate-test-revise iterations equal more reliability. Over 900 three-revision trajectories on HumanEval repairs, current correctness DROPS from 0.820 after one revision to 0.673 after two even as ever-correct rises to 0.847 -- the loop finds the patch an... | Directly attacks the core Loop Engineering assumption that more generate-test-revise iterations equal more reliability. Over 900 three-revision trajectories on HumanEval repairs, current correctness DROPS from 0.820 after one revision to 0.673 after two even as ever-correct rises to 0.847 -- the loop finds the patch an... | Directly attacks the core Loop Engineering assumption that more generate-test-revise iterations equal more reliability. Over 900 three-revision trajectories on HumanEval repairs, current correctness DROPS from 0.820 after one revision to 0.673 after two even as ever-correct rises to 0.847 -- the loop finds the patch an... | State persistence is explicit enough for repeated runs and handoff. Directly attacks the core Loop Engineering assumption that more generate-test-revise iterations equal more reliability. Over 900 three-revision trajectories on HumanEval repairs, current correctness DROPS from 0.820 after one revision to 0.673 after tw... | Use Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair to measure progress and gate completion with repeatable evidence. | Research source arXiv:2607.24604; inspect its method and evaluation before treating results as production evidence. | medium | README.md | 1,048 | https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L1048 | 2026-07-28 | Verify | verify | Gate progress with tests, evals, and evidence. | verification;state | researcher;evaluator | harness | enabling | research-preprint | A | ok | https://arxiv.org/abs/2607.24604 | [2607.24604] Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair | Generate--test--revise loops are common in coding agents, but repetition alone provides no reliability guarantee. We study the gap between finding a correct patch and retaining, verifying, and submitting it. A sealed five-seed study over 30 HumanEval repairs produces 900 three-revision trajectories. Under forced revisi... | Xueping Gao; Jianwei Yang; Qiang Yang | 2026-07-27 | 2026 | arXiv | arXiv | 11 pages, 4 figures, 6 tables | cs.CL | arxiv-api | 2607.24604 | 2026-07-29T08:06:54 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.