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-0101
Official Runtime Guides
official-runtime-guides
Docs
📚
GitHub Copilot for Jira Is Now Generally Available
https://github.blog/changelog/2026-06-25-github-copilot-for-jira-is-now-generally-available/
external
github.blog
General availability of Copilot for Jira: delegate a Jira issue to the Copilot coding agent, monitor session progress inside the issue, and send follow-up instructions that continue the same draft pull request instead of starting a new one.
General availability of Copilot for Jira: delegate a Jira issue to the Copilot coding agent, monitor session progress inside the issue, and send follow-up instructions that continue the same draft pull request instead of starting a new one.
General availability of Copilot for Jira: delegate a Jira issue to the Copilot coding agent, monitor session progress inside the issue, and send follow-up instructions that continue the same draft pull request instead of starting a new one.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. General availability of Copilot for Jira: delegate a Jira issue to the Copilot coding agent, monitor session progress inside the issue, and send follow-up instructions that continue the same draft p...
Use GitHub Copilot for Jira Is Now Generally Available to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
662
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L662
Build
build
Choose runtimes, tools, and delegation surfaces.
intake;delegation
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-06-25-github-copilot-for-jira-is-now-generally-available/
GitHub Copilot for Jira is now generally available - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
GitHub Copilot for Jira is now generally available. Since launching the public preview in March 2026, we have shipped a series of enhancements based on your feedback, including model selection,…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0102
Official Runtime Guides
official-runtime-guides
Docs
📚
Copilot Agent Session Streaming (Public Preview)
https://github.blog/changelog/2026-07-02-copilot-agent-session-streaming-is-now-in-public-preview/
external
github.blog
Public preview that streams Copilot agent session activity, including prompts, responses, and tool calls, from cloud agents, the CLI, and IDEs to SIEM-compatible endpoints and a REST API, giving enterprises an audit trail for delegated agent work.
Public preview that streams Copilot agent session activity, including prompts, responses, and tool calls, from cloud agents, the CLI, and IDEs to SIEM-compatible endpoints and a REST API, giving enterprises an audit trail for delegated agent work.
Public preview that streams Copilot agent session activity, including prompts, responses, and tool calls, from cloud agents, the CLI, and IDEs to SIEM-compatible endpoints and a REST API, giving enterprises an audit trail for delegated agent work.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Public preview that streams Copilot agent session activity, including prompts, responses, and tool calls, from cloud agents, the CLI, and IDEs to SIEM-compatible endpoints and a REST API, giving ent...
Use Copilot Agent Session Streaming (Public Preview) to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
663
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L663
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;delegation
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-07-02-copilot-agent-session-streaming-is-now-in-public-preview/
Copilot agent session streaming is now in public preview - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
GitHub Enterprise Cloud customers with enterprise managed users can now access GitHub Copilot agent session data across all Copilot clients, including: Cloud agents operating on github.com and data resident deployments…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0103
Official Runtime Guides
official-runtime-guides
Docs
📚
Security Reviews in the GitHub Copilot App
https://github.blog/changelog/2026-07-14-security-reviews-now-available-in-the-github-copilot-app
external
github.blog
Changelog adding on-demand security reviews inside the Copilot app, so an agent's proposed changes can be scanned for vulnerabilities before they are merged.
Changelog adding on-demand security reviews inside the Copilot app, so an agent's proposed changes can be scanned for vulnerabilities before they are merged.
Changelog adding on-demand security reviews inside the Copilot app, so an agent's proposed changes can be scanned for vulnerabilities before they are merged.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Changelog adding on-demand security reviews inside the Copilot app, so an agent's proposed changes can be scanned for vulnerabilities before they are merged.
Use Security Reviews in the GitHub Copilot App to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
664
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L664
2026-07-15
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-07-14-security-reviews-now-available-in-the-github-copilot-app/
Security reviews now available in the GitHub Copilot app - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
You can now run a security review on your in-flight code changes directly from the GitHub Copilot app. The /security-review slash command is shipping in public preview, bringing the same…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0104
Official Runtime Guides
official-runtime-guides
Docs
📚
Cursor cloud agents
https://cursor.com/docs/cloud-agent
external
cursor.com
Remote agents that work asynchronously in isolated environments and hand results back for review.
Remote agents that work asynchronously in isolated environments and hand results back for review.
Remote agents that work asynchronously in isolated environments and hand results back for review.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Remote agents that work asynchronously in isolated environments and hand results back for review.
Use Cursor cloud agents to choose an implementation surface for repeatable agent work.
Primary official documentation from cursor.com; use it for current product or standard behavior.
high
README.md
669
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L669
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://cursor.com/docs/cloud-agent
Cloud Agents | Cursor Docs
Run Agent in the cloud for continuous coding assistance.
Cursor Documentation
html-meta
2026-08-07T12:31:05
ale-0105
Official Runtime Guides
official-runtime-guides
Docs
📚
Cursor 3.8: Improvements to Cursor Automations
https://cursor.com/changelog/06-18-26
external
cursor.com
Cursor 3.8 changelog introducing an /automate skill that configures an automation's triggers, instructions, and tools from a plain-language description, plus Slack emoji-reaction and five new GitHub event triggers for dispatching cloud agents.
Cursor 3.8 changelog introducing an /automate skill that configures an automation's triggers, instructions, and tools from a plain-language description, plus Slack emoji-reaction and five new GitHub event triggers for dispatching cloud agents.
Cursor 3.8 changelog introducing an /automate skill that configures an automation's triggers, instructions, and tools from a plain-language description, plus Slack emoji-reaction and five new GitHub event triggers for dispatching cloud agents.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Cursor 3.8 changelog introducing an /automate skill that configures an automation's triggers, instructions, and tools from a plain-language description, plus Slack emoji-reaction and five new GitHub...
Use Cursor 3.8: Improvements to Cursor Automations to choose an implementation surface for repeatable agent work.
Primary official documentation from cursor.com; use it for current product or standard behavior.
high
README.md
670
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L670
Build
build
Choose runtimes, tools, and delegation surfaces.
trigger;workspace
builder
harness
enabling
official-documentation
A
ok
https://cursor.com/changelog/06-18-26
Improvements to Cursor Automations · Cursor
Cursor
html-meta
2026-08-07T12:31:05
ale-0106
Official Runtime Guides
official-runtime-guides
Blog
📝
Expanding Our Long-Running Agents Research Preview
https://cursor.com/blog/long-running-agents
external
cursor.com
Cursor's research preview of days-long autonomous agents gated by an upfront human-approved plan and cross-checked by multiple agents, reporting merge rates comparable to standard agents on runs as large as 52 hours and 151k lines.
Cursor's research preview of days-long autonomous agents gated by an upfront human-approved plan and cross-checked by multiple agents, reporting merge rates comparable to standard agents on runs as large as 52 hours and 151k lines.
Cursor's research preview of days-long autonomous agents gated by an upfront human-approved plan and cross-checked by multiple agents, reporting merge rates comparable to standard agents on runs as large as 52 hours and 151k lines.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Cursor's research preview of days-long autonomous agents gated by an upfront human-approved plan and cross-checked by multiple agents, reporting merge rates comparable to standard agents on runs as ...
Use Expanding Our Long-Running Agents Research Preview to choose an implementation surface for repeatable agent work.
Contextual source from cursor.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
671
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L671
Build
build
Choose runtimes, tools, and delegation surfaces.
escalation
builder
harness
enabling
practitioner-analysis
B
ok
https://cursor.com/blog/long-running-agents
Expanding our long-running agents research preview · Cursor
Long-running agents are now available in the Cursor web app for Ultra, Teams, and Enterprise users.
Cursor Team
Cursor
html-meta
2026-08-07T12:31:05
ale-0107
Official Runtime Guides
official-runtime-guides
Docs
📚
Cursor 3.11: Side Chats, Transcript Search, and Cloud Agent Hooks
https://cursor.com/changelog/side-chat
external
cursor.com
Cursor changelog adding cloud-agent hooks (before-submit, after-response, after-thought, stop, and subagent-start) that the platform pitches for gating and observing background agent runs.
Cursor changelog adding cloud-agent hooks (before-submit, after-response, after-thought, stop, and subagent-start) that the platform pitches for gating and observing background agent runs.
Cursor changelog adding cloud-agent hooks (before-submit, after-response, after-thought, stop, and subagent-start) that the platform pitches for gating and observing background agent runs.
The work separates roles across agents, verifiers, or orchestration layers. Cursor changelog adding cloud-agent hooks (before-submit, after-response, after-thought, stop, and subagent-start) that the platform pitches for gating and observing background agent runs.
Use Cursor 3.11: Side Chats, Transcript Search, and Cloud Agent Hooks to choose an implementation surface for repeatable agent work.
Primary official documentation from cursor.com; use it for current product or standard behavior.
high
README.md
672
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L672
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation;exit
builder
harness
enabling
official-documentation
A
ok
https://cursor.com/changelog/side-chat
Side Chats and Conversation Search · Cursor
Cursor
html-meta
2026-08-07T12:31:05
ale-0108
Official Runtime Guides
official-runtime-guides
Docs
📚
Jules
https://jules.google/docs
external
jules.google
Google's asynchronous coding agent that plans, executes tasks in isolated cloud VMs, and returns reviewable diffs.
Google's asynchronous coding agent that plans, executes tasks in isolated cloud VMs, and returns reviewable diffs.
Google's asynchronous coding agent that plans, executes tasks in isolated cloud VMs, and returns reviewable diffs.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Google's asynchronous coding agent that plans, executes tasks in isolated cloud VMs, and returns reviewable diffs.
Use Jules to choose an implementation surface for repeatable agent work.
Primary official documentation from jules.google; use it for current product or standard behavior.
high
README.md
677
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L677
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://jules.google/docs
jules.google
domain-fallback
2026-08-07T12:31:05
ale-0109
Official Runtime Guides
official-runtime-guides
Docs
📚
Devin Docs
https://docs.devin.ai/get-started/devin-intro
external
docs.devin.ai
Documentation for a long-running autonomous software engineer with sessions, playbooks, knowledge, and review boundaries.
Documentation for a long-running autonomous software engineer with sessions, playbooks, knowledge, and review boundaries.
Documentation for a long-running autonomous software engineer with sessions, playbooks, knowledge, and review boundaries.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Documentation for a long-running autonomous software engineer with sessions, playbooks, knowledge, and review boundaries.
Use Devin Docs to choose an implementation surface for repeatable agent work.
Primary official documentation from docs.devin.ai; use it for current product or standard behavior.
high
README.md
678
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L678
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://docs.devin.ai/get-started/devin-intro
Introducing Devin - Devin Docs
Devin is the AI software engineer, built to help ambitious engineering teams crush their backlogs.
Devin Docs
html-meta
2026-08-07T12:31:05
ale-0110
Official Runtime Guides
official-runtime-guides
Blog
📝
Amp: Agents, Anywhere
https://ampcode.com/news/agents-anywhere
external
ampcode.com
Amp launches remote agent creation on any machine with shell access plus a headless runner mode that lets multiple agents run concurrently without a terminal UI.
Amp launches remote agent creation on any machine with shell access plus a headless runner mode that lets multiple agents run concurrently without a terminal UI.
Amp launches remote agent creation on any machine with shell access plus a headless runner mode that lets multiple agents run concurrently without a terminal UI.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Amp launches remote agent creation on any machine with shell access plus a headless runner mode that lets multiple agents run concurrently without a terminal UI.
Use Amp: Agents, Anywhere to choose an implementation surface for repeatable agent work.
Contextual source from ampcode.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
679
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L679
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
practitioner-analysis
B
ok
https://ampcode.com/news/agents-anywhere
Agents, Anywhere - Amp
Remotely start agents anywhere you can run 'amp'
ampcode.com
domain-fallback
2026-08-07T12:31:05
ale-0111
Official Runtime Guides
official-runtime-guides
Blog
📝
Amp: Right on Schedule
https://ampcode.com/news/schedule
external
ampcode.com
Amp adds scheduled agent runs, letting recurring work fire on a cadence with results reported back, moving the platform from on-demand sessions toward standing loops.
Amp adds scheduled agent runs, letting recurring work fire on a cadence with results reported back, moving the platform from on-demand sessions toward standing loops.
Amp adds scheduled agent runs, letting recurring work fire on a cadence with results reported back, moving the platform from on-demand sessions toward standing loops.
The trigger or cadence is explicit, making the workflow recurring rather than one-off. Amp adds scheduled agent runs, letting recurring work fire on a cadence with results reported back, moving the platform from on-demand sessions toward standing loops.
Use Amp: Right on Schedule to choose an implementation surface for repeatable agent work.
Contextual source from ampcode.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
680
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L680
2026-07-22
Build
build
Choose runtimes, tools, and delegation surfaces.
trigger
builder
harness
enabling
practitioner-analysis
B
ok
https://ampcode.com/news/schedule
Right on Schedule - Amp
Agents can now set their own schedules, wake themselves up, and keep working.
ampcode.com
domain-fallback
2026-08-07T12:31:05
ale-0112
Official Runtime Guides
official-runtime-guides
Docs
📚
Claude Code What's New, Week 29
https://code.claude.com/docs/en/whats-new/2026-w29
external
code.claude.com
Weekly digest of Claude Code changes relevant to recurring agent work, following the Week 28 loop-integrity updates.
Weekly digest of Claude Code changes relevant to recurring agent work, following the Week 28 loop-integrity updates.
Weekly digest of Claude Code changes relevant to recurring agent work, following the Week 28 loop-integrity updates.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Weekly digest of Claude Code changes relevant to recurring agent work, following the Week 28 loop-integrity updates.
Use Claude Code What's New, Week 29 to choose an implementation surface for repeatable agent work.
Primary official documentation from code.claude.com; use it for current product or standard behavior.
high
README.md
681
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L681
2026-07-22
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://code.claude.com/docs/en/whats-new/2026-w29
Week 29 · July 13–17, 2026 - Claude Code Docs
Pull live data into published artifacts through MCP connectors, and use Claude Code with a screen reader in the new screen reader mode.
2026
Claude Code Docs
html-meta
2026-08-07T12:31:05
ale-0113
Official Runtime Guides
official-runtime-guides
Docs
📚
Copilot Code Review: Customization and Configurability
https://github.blog/changelog/2026-07-17-copilot-code-review-customization-and-configurability-improvements
external
github.blog
Changelog adding customization and configurability to Copilot code review, sharpening the automated review gate teams put between agent-written changes and merge.
Changelog adding customization and configurability to Copilot code review, sharpening the automated review gate teams put between agent-written changes and merge.
Changelog adding customization and configurability to Copilot code review, sharpening the automated review gate teams put between agent-written changes and merge.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Changelog adding customization and configurability to Copilot code review, sharpening the automated review gate teams put between agent-written changes and merge.
Use Copilot Code Review: Customization and Configurability to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
682
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L682
2026-07-22
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-07-17-copilot-code-review-customization-and-configurability-improvements/
Copilot code review: Customization and configurability improvements - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
Copilot code review now utilizes a firewall, custom setup steps, and independent runner configurations. It now reads custom instructions from the head branch to allow for easy testing and validation…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0114
Official Runtime Guides
official-runtime-guides
Blog
📝
Amp: Event Driven Orbs
https://ampcode.com/news/event-driven-orbs
external
ampcode.com
Amp's July 23, 2026 launch letting orbs react to outside events: a plugin registers a durable webhook endpoint, each incoming event's signature is verified, and the configured handler runs in an orb thread seeded with trusted event metadata, turning GitHub CI failures, Linear issues, and Discord messages into event-tri...
Amp's July 23, 2026 launch letting orbs react to outside events: a plugin registers a durable webhook endpoint, each incoming event's signature is verified, and the configured handler runs in an orb thread seeded with trusted event metadata, turning GitHub CI failures, Linear issues, and Discord messages into event-tri...
Amp's July 23, 2026 launch letting orbs react to outside events: a plugin registers a durable webhook endpoint, each incoming event's signature is verified, and the configured handler runs in an orb thread seeded with trusted event metadata, turning GitHub CI failures, Linear issues, and Discord messages into event-tri...
Durable execution and replay are treated as first-class loop infrastructure. Amp's July 23, 2026 launch letting orbs react to outside events: a plugin registers a durable webhook endpoint, each incoming event's signature is verified, and the configured handler runs in an orb thread seeded with trusted event metadata, t...
Use Amp: Event Driven Orbs to choose an implementation surface for repeatable agent work.
Contextual source from ampcode.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
683
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L683
2026-07-24
Build
build
Choose runtimes, tools, and delegation surfaces.
trigger;intake;workspace;verification
builder
harness
enabling
practitioner-analysis
B
ok
https://ampcode.com/news/event-driven-orbs
Event Driven Orbs - Amp
Orbs can now receive requests and react to outside events.
ampcode.com
domain-fallback
2026-08-07T12:31:05
ale-0115
Official Runtime Guides
official-runtime-guides
Docs
📚
Scan Your Codebase for Vulnerabilities
https://code.claude.com/docs/en/claude-security
external
code.claude.com
Anthropic's Claude Security plugin runs a multi-agent deep scan of a repository or diff inside a Claude Code session: dynamic workflows orchestrate agents that map the architecture, build a threat model, and hunt vulnerabilities, with independent verifier agents gating every finding before it reaches the report. Revisi...
Anthropic's Claude Security plugin runs a multi-agent deep scan of a repository or diff inside a Claude Code session: dynamic workflows orchestrate agents that map the architecture, build a threat model, and hunt vulnerabilities, with independent verifier agents gating every finding before it reaches the report. Revisi...
Anthropic's Claude Security plugin runs a multi-agent deep scan of a repository or diff inside a Claude Code session: dynamic workflows orchestrate agents that map the architecture, build a threat model, and hunt vulnerabilities, with independent verifier agents gating every finding before it reaches the report. Revisi...
Verification is promoted from a final check to a loop-control signal. Anthropic's Claude Security plugin runs a multi-agent deep scan of a repository or diff inside a Claude Code session: dynamic workflows orchestrate agents that map the architecture, build a threat model, and hunt vulnerabilities, with independent ver...
Use Scan Your Codebase for Vulnerabilities to choose an implementation surface for repeatable agent work.
Primary official documentation from code.claude.com; use it for current product or standard behavior.
high
README.md
684
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L684
2026-07-24
Build
build
Choose runtimes, tools, and delegation surfaces.
delegation;escalation
builder
harness
enabling
official-documentation
A
ok
https://code.claude.com/docs/en/claude-security
Scan your codebase for vulnerabilities - Claude Code Docs
Install the Claude Security plugin to scan your codebase for vulnerabilities in a Claude Code session and turn findings into patches you review and apply.
Claude Code Docs
html-meta
2026-08-07T12:31:05
ale-0116
Official Runtime Guides
official-runtime-guides
Blog
📝
Agent Automation Controls in GitHub Issues
https://github.blog/changelog/2026-07-23-agent-automation-controls-in-github-issues-in-public-preview
external
github.blog
Public preview that puts approval gates on autonomous issue triage: agents driven by Agentic Workflows or the Copilot cloud agent suggest label, field, issue-type, assignee, and close actions with a stated rationale, high-confidence actions apply automatically while lower-confidence ones queue for human accept/decline,...
Public preview that puts approval gates on autonomous issue triage: agents driven by Agentic Workflows or the Copilot cloud agent suggest label, field, issue-type, assignee, and close actions with a stated rationale, high-confidence actions apply automatically while lower-confidence ones queue for human accept/decline,...
Public preview that puts approval gates on autonomous issue triage: agents driven by Agentic Workflows or the Copilot cloud agent suggest label, field, issue-type, assignee, and close actions with a stated rationale, high-confidence actions apply automatically while lower-confidence ones queue for human accept/decline,...
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Public preview that puts approval gates on autonomous issue triage: agents driven by Agentic Workflows or the Copilot cloud agent suggest label, field, issue-type, assignee, and close actions with a...
Use Agent Automation Controls in GitHub Issues to choose an implementation surface for repeatable agent work.
Contextual source from github.blog; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
685
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L685
2026-07-24
Build
build
Choose runtimes, tools, and delegation surfaces.
intake;escalation
builder
harness
enabling
practitioner-analysis
B
ok
https://github.blog/changelog/2026-07-23-agent-automation-controls-in-github-issues-in-public-preview/
Agent automation controls in GitHub Issues in public preview - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
Agent automations increasingly label, type, assign, and close issues for you. GitHub Issues now shows the reason behind each change and lets you review them before they’re applied, so you…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0117
Official Runtime Guides
official-runtime-guides
Blog
📝
Copilot Cloud Agent for Linear Is Now Generally Available
https://github.blog/changelog/2026-07-23-copilot-cloud-agent-for-linear-is-now-generally-available
external
github.blog
General availability of Copilot cloud agent for Linear: assign a Linear issue to the agent, which analyzes it, opens a draft pull request from an ephemeral GitHub Actions environment, streams progress to the Linear activity timeline, and accepts mid-task steering via comments, with model selection, custom agent, and br...
General availability of Copilot cloud agent for Linear: assign a Linear issue to the agent, which analyzes it, opens a draft pull request from an ephemeral GitHub Actions environment, streams progress to the Linear activity timeline, and accepts mid-task steering via comments, with model selection, custom agent, and br...
General availability of Copilot cloud agent for Linear: assign a Linear issue to the agent, which analyzes it, opens a draft pull request from an ephemeral GitHub Actions environment, streams progress to the Linear activity timeline, and accepts mid-task steering via comments, with model selection, custom agent, and br...
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. General availability of Copilot cloud agent for Linear: assign a Linear issue to the agent, which analyzes it, opens a draft pull request from an ephemeral GitHub Actions environment, streams progre...
Use Copilot Cloud Agent for Linear Is Now Generally Available to choose an implementation surface for repeatable agent work.
Contextual source from github.blog; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
686
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L686
2026-07-24
Build
build
Choose runtimes, tools, and delegation surfaces.
intake
builder
harness
enabling
practitioner-analysis
B
ok
https://github.blog/changelog/2026-07-23-copilot-cloud-agent-for-linear-is-now-generally-available/
Copilot cloud agent for Linear is now generally available - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
You can now assign issues in Linear to Copilot cloud agent, our asynchronous, autonomous background agent. When you assign a Linear issue to Copilot, it will: Analyze the issue contents…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0118
Official Runtime Guides
official-runtime-guides
Blog
📝
The New Rules of Context Engineering for Claude 5 Generation Models
https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models
external
claude.com
Official Anthropic post (Jul 24, 2026) by Thariq Shihipar (MTS): Anthropic removed over 80% of Claude Code's system prompt for Opus 5/Fable 5 with no measurable eval loss, and lays out five shifts, rules-to-judgment, example-to-interface tool design, progressive disclosure, concise tool descriptions, and manual-to-auto...
Official Anthropic post (Jul 24, 2026) by Thariq Shihipar (MTS): Anthropic removed over 80% of Claude Code's system prompt for Opus 5/Fable 5 with no measurable eval loss, and lays out five shifts, rules-to-judgment, example-to-interface tool design, progressive disclosure, concise tool descriptions, and manual-to-auto...
Official Anthropic post (Jul 24, 2026) by Thariq Shihipar (MTS): Anthropic removed over 80% of Claude Code's system prompt for Opus 5/Fable 5 with no measurable eval loss, and lays out five shifts, rules-to-judgment, example-to-interface tool design, progressive disclosure, concise tool descriptions, and manual-to-auto...
Primary-source operational guidance rather than commentary. Official Anthropic post (Jul 24, 2026) by Thariq Shihipar (MTS): Anthropic removed over 80% of Claude Code's system prompt for Opus 5/Fable 5 with no measurable eval loss, and lays out five shifts, rules-to-judgment, example-to-interface tool design, progressi...
Use The New Rules of Context Engineering for Claude 5 Generation Models to choose an implementation surface for repeatable agent work.
Contextual source from claude.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
687
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L687
2026-07-25
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;verification
builder
harness
enabling
practitioner-analysis
B
ok
https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models
The new rules of context engineering for Claude 5 generation models | Claude by Anthropic
We removed over 80% of Claude Code's system prompt for more advanced models. How to apply the lessons we learned to your own context engineering in Claude Code and with your own agents.
Claude
html-meta
2026-08-07T12:31:05
ale-0119
Official Runtime Guides
official-runtime-guides
Docs
📚
Claude Cookbook
https://platform.claude.com/cookbook/
external
platform.claude.com
Anthropic's hosted cookbook of runnable recipes for building with Claude, including agent loops with tool use, memory management, context editing, and evaluation harnesses, kept current with each model generation as the reference implementation source.
Anthropic's hosted cookbook of runnable recipes for building with Claude, including agent loops with tool use, memory management, context editing, and evaluation harnesses, kept current with each model generation as the reference implementation source.
Anthropic's hosted cookbook of runnable recipes for building with Claude, including agent loops with tool use, memory management, context editing, and evaluation harnesses, kept current with each model generation as the reference implementation source.
Evaluation data is used as the feedback signal for improving loop behavior. Anthropic's hosted cookbook of runnable recipes for building with Claude, including agent loops with tool use, memory management, context editing, and evaluation harnesses, kept current with each model generation as the reference implementation...
Use Claude Cookbook to choose an implementation surface for repeatable agent work.
Primary official documentation from platform.claude.com; use it for current product or standard behavior.
high
README.md
688
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L688
2026-07-25
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;verification
builder
harness
enabling
official-documentation
A
ok
https://platform.claude.com/cookbook/
Claude Cookbook
Practical guides and code examples for building with Claude. Learn prompting techniques, tool use, multimodal capabilities, and more.
Anthropic
domain-fallback
2026-08-07T12:31:05
ale-0120
Official Runtime Guides
official-runtime-guides
Docs
📚
The 2026-07-28 MCP Specification
https://blog.modelcontextprotocol.io/posts/2026-07-28/
external
blog.modelcontextprotocol.io
Official MCP release announcement (Jul 28, 2026) by lead maintainers David Soria Parra and Den Delimarsky, publishing the 2026-07-28 spec as final after a ten-week release-candidate validation window.
Official MCP release announcement (Jul 28, 2026) by lead maintainers David Soria Parra and Den Delimarsky, publishing the 2026-07-28 spec as final after a ten-week release-candidate validation window.
Official MCP release announcement (Jul 28, 2026) by lead maintainers David Soria Parra and Den Delimarsky, publishing the 2026-07-28 spec as final after a ten-week release-candidate validation window.
Primary-source operational guidance rather than commentary. Official MCP release announcement (Jul 28, 2026) by lead maintainers David Soria Parra and Den Delimarsky, publishing the 2026-07-28 spec as final after a ten-week release-candidate validation window.
Use The 2026-07-28 MCP Specification to choose an implementation surface for repeatable agent work.
Primary official documentation from blog.modelcontextprotocol.io; use it for current product or standard behavior.
high
README.md
689
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L689
2026-07-28
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://blog.modelcontextprotocol.io/posts/2026-07-28/
The 2026-07-28 Specification | Model Context Protocol Blog
The 2026-07-28 Model Context Protocol specification is out, bringing a stateless protocol core, Multi Round-Trip Requests, header-based routing, cacheable list results, authorization hardening, a formal extensions framework, and updated Tier 1 SDKs.
David Soria Parra (Lead Maintainer), Den Delimarsky (Lead Maintainer)
2026-07-28
2026
Model Context Protocol Blog
html-meta
2026-08-07T12:31:05
ale-0121
Official Runtime Guides
official-runtime-guides
Docs
📚
MCP Tasks Extension
https://modelcontextprotocol.io/extensions/tasks/overview
external
modelcontextprotocol.io
Normative documentation for io.modelcontextprotocol/tasks (SEP-2663), promoted out of experimental core into an official extension with the 2026-07-28 spec and redesigned around durable handles instead of session-bound objects.
Normative documentation for io.modelcontextprotocol/tasks (SEP-2663), promoted out of experimental core into an official extension with the 2026-07-28 spec and redesigned around durable handles instead of session-bound objects.
Normative documentation for io.modelcontextprotocol/tasks (SEP-2663), promoted out of experimental core into an official extension with the 2026-07-28 spec and redesigned around durable handles instead of session-bound objects.
Primary-source operational guidance rather than commentary. Normative documentation for io.modelcontextprotocol/tasks (SEP-2663), promoted out of experimental core into an official extension with the 2026-07-28 spec and redesigned around durable handles instead of session-bound objects.
Use MCP Tasks Extension to choose an implementation surface for repeatable agent work.
Primary official documentation from modelcontextprotocol.io; use it for current product or standard behavior.
high
README.md
690
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L690
2026-07-28
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://modelcontextprotocol.io/extensions/tasks/overview
Tasks - Model Context Protocol
Asynchronous task execution for long-running MCP operations
Model Context Protocol
html-meta
2026-08-07T12:31:05
ale-0122
Official Runtime Guides
official-runtime-guides
Docs
📚
Antigravity CLI
https://github.com/google-antigravity/antigravity-cli
external
github.com
Google's Antigravity CLI release of Jul 28, 2026 turns the agent into a programmable component rather than a terminal session.
Google's Antigravity CLI release of Jul 28, 2026 turns the agent into a programmable component rather than a terminal session.
Google's Antigravity CLI release of Jul 28, 2026 turns the agent into a programmable component rather than a terminal session.
Shows how production platforms expose loops through concrete tools, permissions, skills, agents, and automation features. Google's Antigravity CLI release of Jul 28, 2026 turns the agent into a programmable component rather than a terminal session.
Use Antigravity CLI to choose an implementation surface for repeatable agent work.
Primary official documentation from github.com; use it for current product or standard behavior.
high
README.md
691
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L691
2026-07-28
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace;context;delegation;state
builder
harness
enabling
official-documentation
A
ok
https://github.com/google-antigravity/antigravity-cli
GitHub - google-antigravity/antigravity-cli: Antigravity CLI brings the reasoning, execution, and orchestration capabilities of Antigravity agent harness directly into your terminal. · GitHub
Antigravity CLI brings the reasoning, execution, and orchestration capabilities of Antigravity agent harness directly into your terminal. - google-antigravity/antigravity-cli
2026-05-13
2026
google-antigravity/antigravity-cli
GitHub
github-api
google-antigravity/antigravity-cli
1853
166
2026-05-13T17:38:00Z
2026-08-07T09:50:15Z
2026-08-07T12:31:05
ale-0123
Official Runtime Guides
official-runtime-guides
Docs
📚
Enterprise Managed Settings Now Apply to the GitHub Copilot App
https://github.blog/changelog/2026-07-27-enterprise-managed-settings-now-apply-to-the-github-copilot-app
external
github.blog
GitHub changelog (Jul 27, 2026) extending managed-settings.json governance, previously limited to interactive CLI and VS Code clients, to the Copilot cloud agent, the autonomous background worker that opens PRs from issue assignments.
GitHub changelog (Jul 27, 2026) extending managed-settings.json governance, previously limited to interactive CLI and VS Code clients, to the Copilot cloud agent, the autonomous background worker that opens PRs from issue assignments.
GitHub changelog (Jul 27, 2026) extending managed-settings.json governance, previously limited to interactive CLI and VS Code clients, to the Copilot cloud agent, the autonomous background worker that opens PRs from issue assignments.
The contribution is machine-readable and validation-friendly. GitHub changelog (Jul 27, 2026) extending managed-settings.json governance, previously limited to interactive CLI and VS Code clients, to the Copilot cloud agent, the autonomous background worker that opens PRs from issue assignments.
Use Enterprise Managed Settings Now Apply to the GitHub Copilot App to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
692
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L692
2026-07-28
Build
build
Choose runtimes, tools, and delegation surfaces.
intake
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-07-27-enterprise-managed-settings-now-apply-to-the-github-copilot-app/
Enterprise managed settings in the GitHub Copilot app and Copilot cloud agent - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
You can now govern the GitHub Copilot app and Copilot cloud agent with enterprise managed settings, the same centrally managed policies you use to control Copilot across your enterprise. With…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0124
Official Runtime Guides
official-runtime-guides
Docs
📚
GitHub Copilot in Visual Studio Code, July 2026 Releases
https://github.blog/changelog/2026-07-30-github-copilot-in-visual-studio-code-july-2026-releases/
external
github.blog
The Agents window (public preview) gets the observability and isolation primitives that parallel agent operation needs. Sessions can now be started in a Git worktree, and notably for Copilot, Claude, or Codex sessions, not just Copilot's own, giving each concurrent agent its own checkout instead of contending over one ...
The Agents window (public preview) gets the observability and isolation primitives that parallel agent operation needs. Sessions can now be started in a Git worktree, and notably for Copilot, Claude, or Codex sessions, not just Copilot's own, giving each concurrent agent its own checkout instead of contending over one ...
The Agents window (public preview) gets the observability and isolation primitives that parallel agent operation needs. Sessions can now be started in a Git worktree, and notably for Copilot, Claude, or Codex sessions, not just Copilot's own, giving each concurrent agent its own checkout instead of contending over one ...
Workspace isolation is part of the loop design, not an afterthought. The Agents window (public preview) gets the observability and isolation primitives that parallel agent operation needs. Sessions can now be started in a Git worktree, and notably for Copilot, Claude, or Codex sessions, not just Copilot's own, giving e...
Use GitHub Copilot in Visual Studio Code, July 2026 Releases to choose an implementation surface for repeatable agent work.
Primary official documentation from github.blog; use it for current product or standard behavior.
high
README.md
693
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L693
2026-08-02
Build
build
Choose runtimes, tools, and delegation surfaces.
workspace
builder
harness
enabling
official-documentation
A
ok
https://github.blog/changelog/2026-07-30-github-copilot-in-visual-studio-code-july-2026-releases/
GitHub Copilot in Visual Studio Code, July 2026 releases - GitHub Changelog LinkedIn icon Instagram icon YouTube icon X icon TikTok icon Twitch icon GitHub icon
This changelog covers VS Code v1.127 through v1.131, shipped throughout July 2026. These releases improve how you work with agents, review changes, use chat, and navigate VS Code. They also…
2026
The GitHub Blog
html-meta
2026-08-07T12:31:05
ale-0125
Research Foundations
research-foundations
Paper
📄
ReAct: Synergizing Reasoning and Acting in Language Models
https://arxiv.org/abs/2210.03629
external
arxiv.org
Foundational reason-act-observe loop for tool-using language agents.
Foundational reason-act-observe loop for tool-using language agents.
Foundational reason-act-observe loop for tool-using language agents.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Foundational reason-act-observe loop for tool-using language agents.
Use ReAct: Synergizing Reasoning and Acting in Language Models to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2210.03629; inspect its method and evaluation before treating results as production evidence.
medium
README.md
701
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L701
Learn
learn
Understand the field and its boundaries.
workspace
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://openreview.net/forum?id=WE_vluYUL-X
[2210.03629] ReAct: Synergizing Reasoning and Acting in Language Models
While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision making, their abilities for reasoning (e.g. chain-of-thought prompting) and acting (e.g. action plan generation) have primarily been studied as separate topics. In this paper, we ...
Shunyu Yao; Jeffrey Zhao; Dian Yu; Nan Du; Izhak Shafran; Karthik Narasimhan; Yuan Cao
2023
2023
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published in International Conference on Learning Representations (ICLR); the linked arXiv record remains available for open access.
cs.CL
OpenReview proceedings record
2210.03629
2026-08-07T12:31:05
ale-0126
Research Foundations
research-foundations
Paper
📄
Reflexion: Language Agents with Verbal Reinforcement Learning
https://arxiv.org/abs/2303.11366
external
arxiv.org
Converts environment feedback into written reflections stored in memory for future attempts.
Converts environment feedback into written reflections stored in memory for future attempts.
Converts environment feedback into written reflections stored in memory for future attempts.
Persistent memory is treated as an external runtime artifact. Converts environment feedback into written reflections stored in memory for future attempts.
Use Reflexion: Language Agents with Verbal Reinforcement Learning to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2303.11366; inspect its method and evaluation before treating results as production evidence.
medium
README.md
702
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L702
Learn
learn
Understand the field and its boundaries.
context
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.neurips.cc/paper_files/paper/2023/hash/1b44b878bb782e6954cd888628510e90-Abstract-Conference.html
[2303.11366] Reflexion: Language Agents with Verbal Reinforcement Learning
Large language models (LLMs) have been increasingly used to interact with external environments (e.g., games, compilers, APIs) as goal-driven agents. However, it remains challenging for these language agents to quickly and efficiently learn from trial-and-error as traditional reinforcement learning methods require exte...
Noah Shinn; Federico Cassano; Edward Berman; Ashwin Gopinath; Karthik Narasimhan; Shunyu Yao
2023
2023
Advances in Neural Information Processing Systems 36 (NeurIPS)
Neural Information Processing Systems Foundation
10.52202/075280-0377
Published in Advances in Neural Information Processing Systems 36 (NeurIPS); the linked arXiv record remains available for open access.
cs.AI
NeurIPS proceedings and DOI records
2303.11366
2026-08-07T12:31:05
ale-0127
Research Foundations
research-foundations
Paper
📄
Self-Refine: Iterative Refinement with Self-Feedback
https://arxiv.org/abs/2303.17651
external
arxiv.org
Generate-feedback-refine loop where a model improves outputs over repeated passes.
Generate-feedback-refine loop where a model improves outputs over repeated passes.
Generate-feedback-refine loop where a model improves outputs over repeated passes.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Generate-feedback-refine loop where a model improves outputs over repeated passes.
Use Self-Refine: Iterative Refinement with Self-Feedback to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2303.17651; inspect its method and evaluation before treating results as production evidence.
medium
README.md
703
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L703
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.neurips.cc/paper_files/paper/2023/hash/91edff07232fb1b55a505a9e9f6c0ff3-Abstract-Conference.html
[2303.17651] Self-Refine: Iterative Refinement with Self-Feedback
Like humans, large language models (LLMs) do not always generate the best output on their first try. Motivated by how humans refine their written text, we introduce Self-Refine, an approach for improving initial outputs from LLMs through iterative feedback and refinement. The main idea is to generate an initial output ...
Aman Madaan; Niket Tandon; Prakhar Gupta; Skyler Hallinan; Luyu Gao; Sarah Wiegreffe; Uri Alon; Nouha Dziri; Shrimai Prabhumoye; Yiming Yang; Shashank Gupta; Bodhisattwa Prasad Majumder; Katherine Hermann; Sean Welleck; Amir Yazdanbakhsh; Peter Clark
2023
2023
Advances in Neural Information Processing Systems 36 (NeurIPS)
Neural Information Processing Systems Foundation
Published in Advances in Neural Information Processing Systems 36 (NeurIPS); the linked arXiv record remains available for open access.
cs.CL
NeurIPS proceedings record
2303.17651
2026-08-07T12:31:05
ale-0128
Research Foundations
research-foundations
Paper
📄
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
https://arxiv.org/abs/2305.11738
external
arxiv.org
Uses tools to ground critique and correction rather than relying only on introspection.
Uses tools to ground critique and correction rather than relying only on introspection.
Uses tools to ground critique and correction rather than relying only on introspection.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Uses tools to ground critique and correction rather than relying only on introspection.
Use CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2305.11738; inspect its method and evaluation before treating results as production evidence.
medium
README.md
704
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L704
Learn
learn
Understand the field and its boundaries.
workspace;verification
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.iclr.cc/paper_files/paper/2024/hash/fef126561bbf9d4467dbb8d27334b8fe-Abstract-Conference.html
[2305.11738] CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
Recent developments in large language models (LLMs) have been impressive. However, these models sometimes show inconsistencies and problematic behavior, such as hallucinating facts, generating flawed code, or creating offensive and toxic content. Unlike these models, humans typically utilize external tools to cross-che...
Zhibin Gou; Zhihong Shao; Yeyun Gong; Yelong Shen; Yujiu Yang; Nan Duan; Weizhu Chen
2024
2024
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published in International Conference on Learning Representations (ICLR); the linked arXiv record remains available for open access.
cs.CL
ICLR proceedings record
2305.11738
2026-08-07T12:31:05
ale-0129
Research Foundations
research-foundations
Paper
📄
Tree of Thoughts
https://arxiv.org/abs/2305.10601
external
arxiv.org
Search over multiple reasoning branches; relevant when loop design needs exploration before committing.
Search over multiple reasoning branches; relevant when loop design needs exploration before committing.
Search over multiple reasoning branches; relevant when loop design needs exploration before committing.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Search over multiple reasoning branches; relevant when loop design needs exploration before committing.
Use Tree of Thoughts to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2305.10601; inspect its method and evaluation before treating results as production evidence.
medium
README.md
705
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L705
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.neurips.cc/paper_files/paper/2023/hash/271db9922b8d1f4dd7aaef84ed5ac703-Abstract.html
[2305.10601] Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Language models are increasingly being deployed for general problem solving across a wide range of tasks, but are still confined to token-level, left-to-right decision-making processes during inference. This means they can fall short in tasks that require exploration, strategic lookahead, or where initial decisions pla...
Shunyu Yao; Dian Yu; Jeffrey Zhao; Izhak Shafran; Thomas L. Griffiths; Yuan Cao; Karthik Narasimhan
2023
2023
Advances in Neural Information Processing Systems 36 (NeurIPS)
Neural Information Processing Systems Foundation
Published in Advances in Neural Information Processing Systems 36 (NeurIPS); the linked arXiv record remains available for open access.
cs.CL
NeurIPS proceedings record
2305.10601
2026-08-07T12:31:05
ale-0130
Research Foundations
research-foundations
Paper
📄
Graph of Thoughts
https://arxiv.org/abs/2308.09687
external
arxiv.org
Generalizes thought structures beyond chains and trees, useful for complex loop planning and aggregation.
Generalizes thought structures beyond chains and trees, useful for complex loop planning and aggregation.
Generalizes thought structures beyond chains and trees, useful for complex loop planning and aggregation.
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Generalizes thought structures beyond chains and trees, useful for complex loop planning and aggregation.
Use Graph of Thoughts to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2308.09687; inspect its method and evaluation before treating results as production evidence.
medium
README.md
706
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L706
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://ojs.aaai.org/index.php/AAAI/article/view/29720
[2308.09687] Graph of Thoughts: Solving Elaborate Problems with Large Language Models
We introduce Graph of Thoughts (GoT): a framework that advances prompting capabilities in large language models (LLMs) beyond those offered by paradigms such as Chain-of-Thought or Tree of Thoughts (ToT). The key idea and primary advantage of GoT is the ability to model the information generated by an LLM as an arbitra...
Maciej Besta; Nils Blach; Ales Kubicek; Robert Gerstenberger; Michal Podstawski; Lukas Gianinazzi; Joanna Gajda; Tomasz Lehmann; Hubert Niewiadomski; Piotr Nyczyk; Torsten Hoefler
2024-03-24
2024
Proceedings of the AAAI Conference on Artificial Intelligence 38 (AAAI)
Association for the Advancement of Artificial Intelligence (AAAI)
10.1609/aaai.v38i16.29720
Published in Proceedings of the AAAI Conference on Artificial Intelligence 38 (AAAI); the linked arXiv record remains available for open access.
cs.CL
AAAI proceedings and DOI records
2308.09687
2026-08-07T12:31:05
ale-0131
Research Foundations
research-foundations
Paper
📄
Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models
https://arxiv.org/abs/2310.04406
external
arxiv.org
Combines search, action, and environment feedback for language agents.
Combines search, action, and environment feedback for language agents.
Combines search, action, and environment feedback for language agents.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Combines search, action, and environment feedback for language agents.
Use Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2310.04406; inspect its method and evaluation before treating results as production evidence.
medium
README.md
707
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L707
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.mlr.press/v235/zhou24r.html
[2310.04406] Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models
While language models (LMs) have shown potential across a range of decision-making tasks, their reliance on simple acting processes limits their broad deployment as autonomous agents. In this paper, we introduce Language Agent Tree Search (LATS) -- the first general framework that synergizes the capabilities of LMs in ...
Andy Zhou; Kai Yan; Michal Shlapentokh-Rothman; Haohan Wang; Yu-Xiong Wang
2024
2024
Proceedings of the 41st International Conference on Machine Learning (ICML)
PMLR
Published in Proceedings of the 41st International Conference on Machine Learning (ICML); the linked arXiv record remains available for open access.
cs.AI
PMLR proceedings record
2310.04406
2026-08-07T12:31:05
ale-0132
Research Foundations
research-foundations
Paper
📄
Voyager: An Open-Ended Embodied Agent with Large Language Models
https://arxiv.org/abs/2305.16291
external
arxiv.org
Demonstrates lifelong skill acquisition through iterative exploration, feedback, and a skill library.
Demonstrates lifelong skill acquisition through iterative exploration, feedback, and a skill library.
Demonstrates lifelong skill acquisition through iterative exploration, feedback, and a skill library.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Demonstrates lifelong skill acquisition through iterative exploration, feedback, and a skill library.
Use Voyager: An Open-Ended Embodied Agent with Large Language Models to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2305.16291; inspect its method and evaluation before treating results as production evidence.
medium
README.md
708
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L708
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://openreview.net/forum?id=ehfRiF0R3a
[2305.16291] Voyager: An Open-Ended Embodied Agent with Large Language Models
We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-gro...
Guanzhi Wang; Yuqi Xie; Yunfan Jiang; Ajay Mandlekar; Chaowei Xiao; Yuke Zhu; Linxi Fan; Anima Anandkumar
2024
2024
Transactions on Machine Learning Research (TMLR)
OpenReview
Published in Transactions on Machine Learning Research (TMLR); the linked arXiv record remains available for open access.
cs.AI
TMLR OpenReview record
2305.16291
2026-08-07T12:31:05
ale-0133
Research Foundations
research-foundations
Paper
📄
Generative Agents: Interactive Simulacra of Human Behavior
https://arxiv.org/abs/2304.03442
external
arxiv.org
Introduces reflection and memory mechanisms for long-running agent behavior.
Introduces reflection and memory mechanisms for long-running agent behavior.
Introduces reflection and memory mechanisms for long-running agent behavior.
Persistent memory is treated as an external runtime artifact. Introduces reflection and memory mechanisms for long-running agent behavior.
Use Generative Agents: Interactive Simulacra of Human Behavior to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2304.03442; inspect its method and evaluation before treating results as production evidence.
medium
README.md
709
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L709
Learn
learn
Understand the field and its boundaries.
context;escalation
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://doi.org/10.1145/3586183.3606763
[2304.03442] Generative Agents: Interactive Simulacra of Human Behavior
Believable proxies of human behavior can empower interactive applications ranging from immersive environments to rehearsal spaces for interpersonal communication to prototyping tools. In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents ...
Joon Sung Park; Joseph C. O'Brien; Carrie J. Cai; Meredith Ringel Morris; Percy Liang; Michael S. Bernstein
2023-10-29
2023
Proceedings of the 36th ACM Symposium on User Interface Software and Technology (UIST)
Association for Computing Machinery
10.1145/3586183.3606763
Published in Proceedings of the 36th ACM Symposium on User Interface Software and Technology (UIST); the linked arXiv record remains available for open access.
cs.HC
ACM DOI record
2304.03442
2026-08-07T12:31:05
ale-0134
Research Foundations
research-foundations
Paper
📄
Measuring AI Ability to Complete Long Software Tasks
https://arxiv.org/abs/2503.14499
external
arxiv.org
METR's task-length time horizon metric; grounds why loop budgets, checkpoints, and escalation matter as autonomous work gets longer.
METR's task-length time horizon metric; grounds why loop budgets, checkpoints, and escalation matter as autonomous work gets longer.
METR's task-length time horizon metric; grounds why loop budgets, checkpoints, and escalation matter as autonomous work gets longer.
Checkpointed state makes long-running agent work recoverable across failures. METR's task-length time horizon metric; grounds why loop budgets, checkpoints, and escalation matter as autonomous work gets longer.
Use Measuring AI Ability to Complete Long Software Tasks to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2503.14499; inspect its method and evaluation before treating results as production evidence.
medium
README.md
710
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L710
Learn
learn
Understand the field and its boundaries.
state;budget;escalation
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://proceedings.neurips.cc/paper_files/paper/2025/hash/85069585133c4c168c865e65d72e9775-Abstract-Conference.html
[2503.14499] Measuring AI Ability to Complete Long Software Tasks
Despite rapid progress on AI benchmarks, the real-world meaning of benchmark performance remains unclear. To quantify the capabilities of AI systems in terms of human capabilities, we propose a new metric: 50%-task-completion time horizon. This is the time humans typically take to complete tasks that AI models can comp...
Thomas Kwa; Ben West; Joel Becker; Amy Deng; Katharyn Garcia; Max Hasin; Sami Jawhar; Megan Kinniment; Nate Rush; Sydney Von Arx; Ryan Bloom; Thomas Broadley; Haoxing Du; Brian Goodrich; Nikola Jurkovic; Luke Harold Miles; Seraphina Nix; Tao Lin; Chris Painter; Neev Parikh; David Rein; Lucas Jun Koba Sato; Hjalmar Wijk...
2025
2025
Advances in Neural Information Processing Systems 38 (NeurIPS)
Neural Information Processing Systems Foundation
Published in Advances in Neural Information Processing Systems 38 (NeurIPS); the linked arXiv record remains available for open access.
cs.AI
NeurIPS proceedings record
2503.14499
2026-08-07T12:31:05
ale-0135
Research Foundations
research-foundations
Blog
📝
Measuring AI Ability to Complete Long Tasks
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
external
metr.org
Accessible summary of the 50% task-completion time horizon and its doubling trend.
Accessible summary of the 50% task-completion time horizon and its doubling trend.
Accessible summary of the 50% task-completion time horizon and its doubling trend.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Accessible summary of the 50% task-completion time horizon and its doubling trend.
Use Measuring AI Ability to Complete Long Tasks to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Contextual source from metr.org; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
711
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L711
Learn
learn
Understand the field and its boundaries.
exit
builder
cross-layer
enabling
practitioner-analysis
B
ok
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
Measuring AI Ability to Complete Long Software Tasks - METR
We propose measuring AI performance in terms of the *length* of tasks AI agents can complete. We show that this metric has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months. Extrapolating this trend predicts that, in under a decade, we will see AI agents that can ...
2025-03-19
2025
METR Blog
metr.org
html-meta
2026-08-07T12:31:05
ale-0136
Research Foundations
research-foundations
Paper
📄
Reflection-Driven Control for Trustworthy Code Agents
https://arxiv.org/abs/2512.21354
external
arxiv.org
Elevates reflection from an external pass to an internal control loop that monitors the agent's decision path during generation and constrains risky steps with low overhead.
Elevates reflection from an external pass to an internal control loop that monitors the agent's decision path during generation and constrains risky steps with low overhead.
Elevates reflection from an external pass to an internal control loop that monitors the agent's decision path during generation and constrains risky steps with low overhead.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Elevates reflection from an external pass to an internal control loop that monitors the agent's decision path during generation and constrains risky steps with low overhead.
Use Reflection-Driven Control for Trustworthy Code Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2512.21354; inspect its method and evaluation before treating results as production evidence.
medium
README.md
712
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L712
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://openreview.net/forum?id=vUtz66IHD1
[2512.21354] Reflection-Driven Control for Trustworthy Code Agents
Contemporary large language model (LLM) agents are remarkably capable, but they still lack reliable safety controls and can produce unconstrained, unpredictable, and even actively harmful outputs. To address this, we introduce Reflection-Driven Control, a standardized and pluggable control module that can be seamlessly...
Bin Wang; Jiazheng Quan; Xingrui Yu; Hansen Hu; Yuhao; Ivor Tsang
2026
2026
AAAI Workshop on Trust and Control in Agentic AI (TrustAgent)
Association for the Advancement of Artificial Intelligence (AAAI)
Published in AAAI Workshop on Trust and Control in Agentic AI (TrustAgent); the linked arXiv record remains available for open access.
cs.CR
AAAI workshop OpenReview record
2512.21354
2026-08-07T12:31:05
ale-0137
Research Foundations
research-foundations
Paper
📄
Hyperagents
https://arxiv.org/abs/2603.19461
external
arxiv.org
Self-referential agents that fold task-solving and self-modification into editable programs, extending the Darwin Godel Machine toward open-ended self-improvement, the loop where an agent rewrites its own improvement mechanism across runs.
Self-referential agents that fold task-solving and self-modification into editable programs, extending the Darwin Godel Machine toward open-ended self-improvement, the loop where an agent rewrites its own improvement mechanism across runs.
Self-referential agents that fold task-solving and self-modification into editable programs, extending the Darwin Godel Machine toward open-ended self-improvement, the loop where an agent rewrites its own improvement mechanism across runs.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Self-referential agents that fold task-solving and self-modification into editable programs, extending the Darwin Godel Machine toward open-ended self-improvement, the loop where an agent rewrites its own improvement mechanism across runs.
Use Hyperagents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2603.19461; inspect its method and evaluation before treating results as production evidence.
medium
README.md
713
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L713
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2603.19461
[2603.19461] Hyperagents
Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to self-improvement rely on fixed, handcrafted meta-level mechanisms, fundamentally limiting how fast such systems can improve. The Darwin G\"odel Machine (D...
Jenny Zhang; Bingchen Zhao; Wannan Yang; Jakob Foerster; Jeff Clune; Minqi Jiang; Sam Devlin; Tatiana Shavrina
2026-03-19
2026
arXiv
arXiv
Code at https://github.com/facebookresearch/Hyperagents
cs.AI
arxiv-api
2603.19461
2026-08-07T12:31:05
ale-0138
Research Foundations
research-foundations
Paper
📄
PARC: An Autonomous Self-Reflective Coding Agent for Robust Execution of Long-Horizon Tasks
https://arxiv.org/abs/2512.03549
external
arxiv.org
Hierarchical plan-execute-assess loops that detect and correct strategic errors during multi-hour autonomous runs.
Hierarchical plan-execute-assess loops that detect and correct strategic errors during multi-hour autonomous runs.
Hierarchical plan-execute-assess loops that detect and correct strategic errors during multi-hour autonomous runs.
The work targets tasks that exceed a single context window or prompt session. Hierarchical plan-execute-assess loops that detect and correct strategic errors during multi-hour autonomous runs.
Use PARC: An Autonomous Self-Reflective Coding Agent for Robust Execution of Long-Horizon Tasks to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2512.03549; inspect its method and evaluation before treating results as production evidence.
medium
README.md
714
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L714
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2512.03549
[2512.03549] PARC: An Autonomous Self-Reflective Coding Agent for Robust Execution of Long-Horizon Tasks
We introduce PARC, a coding agent for the autonomous and robust execution of long-horizon computational tasks. PARC is built on a hierarchical multi-agent architecture incorporating task planning, execution, and a mechanism that evaluates its own actions and their outcomes from an independent context and provides feedb...
Yuki Orimo; Iori Kurata; Hodaka Mori; Ryuhei Okuno; Ryohto Sawada; Daisuke Okanohara
2025-12-03
2025
arXiv
arXiv
cs.AI
arxiv-api
2512.03549
2026-08-07T12:31:05
ale-0139
Research Foundations
research-foundations
Paper
📄
When the Specification Emerges: Benchmarking Faithfulness Loss in Long-Horizon Coding Agents
https://arxiv.org/abs/2603.17104
external
arxiv.org
Measures how agents drift from intent when specifications arrive incrementally across a long loop, and proposes a mitigation that recovers most of the loss.
Measures how agents drift from intent when specifications arrive incrementally across a long loop, and proposes a mitigation that recovers most of the loss.
Measures how agents drift from intent when specifications arrive incrementally across a long loop, and proposes a mitigation that recovers most of the loss.
The work targets tasks that exceed a single context window or prompt session. Measures how agents drift from intent when specifications arrive incrementally across a long loop, and proposes a mitigation that recovers most of the loss.
Use When the Specification Emerges: Benchmarking Faithfulness Loss in Long-Horizon Coding Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2603.17104; inspect its method and evaluation before treating results as production evidence.
medium
README.md
715
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L715
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2603.17104
[2603.17104] When the Specification Emerges: Benchmarking Faithfulness Loss in Long-Horizon Coding Agents
Current coding-agent benchmarks usually pro- vide the full task specification upfront. Real research coding often does not: the intended system is progressively disclosed through in- teraction, requiring the agent to track durable design commitments across a long session. We introduce a benchmark for this setting and s...
Lu Yan; Xuan Chen; Xiangyu Zhang
2026-03-17
2026
arXiv
arXiv
cs.SE
arxiv-api
2603.17104
2026-08-07T12:31:05
ale-0140
Research Foundations
research-foundations
Tool
🧰
Reflexion code
https://github.com/noahshinn/reflexion
external
github.com
Reference implementation and experiments for verbal reinforcement loops.
Reference implementation and experiments for verbal reinforcement loops.
Reference implementation and experiments for verbal reinforcement loops.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Reference implementation and experiments for verbal reinforcement loops.
Use Reflexion code to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Inspectable GitHub source (3,223 stars; 314 forks; MIT license; updated 2026-08-06); popularity is context, not proof of reliability.
medium
README.md
716
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L716
Learn
learn
Understand the field and its boundaries.
whole-loop
builder
cross-layer
enabling
source-implementation
A
ok
https://github.com/noahshinn/reflexion
GitHub - noahshinn/reflexion: [NeurIPS 2023] Reflexion: Language Agents with Verbal Reinforcement Learning · GitHub
[NeurIPS 2023] Reflexion: Language Agents with Verbal Reinforcement Learning - noahshinn/reflexion
2023-03-22
2023
noahshinn/reflexion
GitHub
github-api
noahshinn/reflexion
3223
314
MIT
2023-03-22T06:38:53Z
2026-08-06T11:36:01Z
2026-08-07T12:31:05
ale-0141
Research Foundations
research-foundations
Paper
📄
Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting
https://arxiv.org/abs/2607.00038
external
arxiv.org
Position paper that formalizes the loop specification (trigger, goal, verification step, stopping rule, memory) as a reusable artifact handed to an agent harness, with a taxonomy, a five-level verification ladder, and a hand-coded analysis of fifty real-world loops.
Position paper that formalizes the loop specification (trigger, goal, verification step, stopping rule, memory) as a reusable artifact handed to an agent harness, with a taxonomy, a five-level verification ladder, and a hand-coded analysis of fifty real-world loops.
Position paper that formalizes the loop specification (trigger, goal, verification step, stopping rule, memory) as a reusable artifact handed to an agent harness, with a taxonomy, a five-level verification ladder, and a hand-coded analysis of fifty real-world loops.
Verification is promoted from a final check to a loop-control signal. Position paper that formalizes the loop specification (trigger, goal, verification step, stopping rule, memory) as a reusable artifact handed to an agent harness, with a taxonomy, a five-level verification ladder, and a hand-coded analysis of fifty r...
Use Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.00038; inspect its method and evaluation before treating results as production evidence.
medium
README.md
717
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L717
Learn
learn
Understand the field and its boundaries.
objective;trigger;context;verification;exit
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.00038
[2607.00038] Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting
In mid-2026 a slogan reorganized how practitioners talk about coding agents: stop prompting your agent, start designing the loop that prompts it. We take this claim seriously and give it a careful treatment. We call the object of the new practice the loop specification: a bounded, reusable artifact, made of a trigger, ...
Sandeco Macedo
2026-06-28
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.00038
2026-08-07T12:31:05
ale-0142
Research Foundations
research-foundations
Paper
📄
From Question Answering to Task Completion: A Survey on Agent System and Harness Design
https://arxiv.org/abs/2606.20683
external
arxiv.org
Survey that decomposes the agent execution harness into six runtime responsibilities (observation, context, control, action, state, verification) and argues task performance emerges from the interaction of model, runtime, task structure, and evaluation rather than the model alone.
Survey that decomposes the agent execution harness into six runtime responsibilities (observation, context, control, action, state, verification) and argues task performance emerges from the interaction of model, runtime, task structure, and evaluation rather than the model alone.
Survey that decomposes the agent execution harness into six runtime responsibilities (observation, context, control, action, state, verification) and argues task performance emerges from the interaction of model, runtime, task structure, and evaluation rather than the model alone.
Verification is promoted from a final check to a loop-control signal. Survey that decomposes the agent execution harness into six runtime responsibilities (observation, context, control, action, state, verification) and argues task performance emerges from the interaction of model, runtime, task structure, and evaluati...
Use From Question Answering to Task Completion: A Survey on Agent System and Harness Design to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2606.20683; inspect its method and evaluation before treating results as production evidence.
medium
README.md
718
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L718
Learn
learn
Understand the field and its boundaries.
context;verification;state;exit
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2606.20683
[2606.20683] From Question Answering to Task Completion: A Survey on Agent System and Harness Design
LLM-based agents mark a shift from passive question answering to active task completion: they perceive environments, invoke tools, maintain state, and act over extended horizons. As agent systems have evolved from prompt engineering to workflows and context engineering, harness engineering, and agent-native training wi...
Jianyuan Guo; Zhiwei Hao; Chengcheng Wang; Cheng Fan; Tingzhang Luo; Hongguang Li; Ying Gao; Hefei Mei; Jiankun Peng; Rongjian Xu; Minjing Dong; Han Wu; Mengyu Zheng; Kai Han; Shiqi Wang; Chang Xu; Yunhe Wang
2026-06-14
2026
arXiv
arXiv
cs.AI
arxiv-api
2606.20683
2026-08-07T12:31:05
ale-0143
Research Foundations
research-foundations
Paper
📄
MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems
https://arxiv.org/abs/2605.22794
external
arxiv.org
Self-evolution loop where the agent rewrites its own source code, with each change anchored to a production failure and accepted only after deterministic replay verification with rollback, lifting a four-task mean grader score from 0.25 to 0.61 without human intervention.
Self-evolution loop where the agent rewrites its own source code, with each change anchored to a production failure and accepted only after deterministic replay verification with rollback, lifting a four-task mean grader score from 0.25 to 0.61 without human intervention.
Self-evolution loop where the agent rewrites its own source code, with each change anchored to a production failure and accepted only after deterministic replay verification with rollback, lifting a four-task mean grader score from 0.25 to 0.61 without human intervention.
Durable execution and replay are treated as first-class loop infrastructure. Self-evolution loop where the agent rewrites its own source code, with each change anchored to a production failure and accepted only after deterministic replay verification with rollback, lifting a four-task mean grader score from 0.25 to 0.6...
Use MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2605.22794; inspect its method and evaluation before treating results as production evidence.
medium
README.md
719
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L719
Learn
learn
Understand the field and its boundaries.
verification;state;escalation
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2605.22794
[2605.22794] MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems
Autonomous agentic systems are largely static after deployment: they do not learn from user interactions, and recurring failures persist until the next human-driven update ships a fix. Self-evolving agents have emerged in response, but all confine evolution to text-mutable artifacts -- skill files, prompt configuration...
Qianshu Cai; Yonggang Zhang; Xianzhang Jia; Huajiang Zheng; Wei Xue; Jun Song; Xinmei Tian; Yike Guo
2026-05-21
2026
arXiv
arXiv
12 pages, 3 figures, 2 tables. Preprint. Code: https://github.com/hkgai-official/Moss
cs.AI
arxiv-api
2605.22794
2026-08-07T12:31:05
ale-0144
Research Foundations
research-foundations
Blog
📝
METR Time Horizon 1.1
https://metr.org/blog/2026-1-29-time-horizon-1-1/
external
metr.org
Update to METR's time-horizon methodology, expanding the task suite to 228 tasks (31 at 8+ hours), migrating to the open-source Inspect framework, and revising the post-2023 capability doubling time to roughly 131 days.
Update to METR's time-horizon methodology, expanding the task suite to 228 tasks (31 at 8+ hours), migrating to the open-source Inspect framework, and revising the post-2023 capability doubling time to roughly 131 days.
Update to METR's time-horizon methodology, expanding the task suite to 228 tasks (31 at 8+ hours), migrating to the open-source Inspect framework, and revising the post-2023 capability doubling time to roughly 131 days.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Update to METR's time-horizon methodology, expanding the task suite to 228 tasks (31 at 8+ hours), migrating to the open-source Inspect framework, and revising the post-2023 capability doubling time to roughly 131 days.
Use METR Time Horizon 1.1 to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Contextual source from metr.org; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
720
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L720
Learn
learn
Understand the field and its boundaries.
whole-loop
builder
cross-layer
enabling
practitioner-analysis
B
ok
https://metr.org/blog/2026-1-29-time-horizon-1-1/
Time Horizon 1.1 - METR Substack twitter Bluesky
We’re releasing a new version of our time horizon estimates (TH1.1), using more tasks and a new eval infrastructure.
2026-01-29
2026
METR Blog
metr.org
html-meta
2026-08-07T12:31:05
ale-0145
Research Foundations
research-foundations
Paper
📄
MetaSkill-Evolve: Recursive Self-Improvement via Two-Timescale Meta-Skill Evolution
https://arxiv.org/abs/2607.05297
external
arxiv.org
Two-timescale recursive self-improvement where a fast loop rewrites task skills from execution traces while a slow loop evolves the meta-skill governing improvement itself, gaining up to 23.5 points on OfficeQA, SealQA, and ALFWorld.
Two-timescale recursive self-improvement where a fast loop rewrites task skills from execution traces while a slow loop evolves the meta-skill governing improvement itself, gaining up to 23.5 points on OfficeQA, SealQA, and ALFWorld.
Two-timescale recursive self-improvement where a fast loop rewrites task skills from execution traces while a slow loop evolves the meta-skill governing improvement itself, gaining up to 23.5 points on OfficeQA, SealQA, and ALFWorld.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Two-timescale recursive self-improvement where a fast loop rewrites task skills from execution traces while a slow loop evolves the meta-skill governing improvement itself, gaining up to 23.5 points on OfficeQA, SealQA, and ALFWorld.
Use MetaSkill-Evolve: Recursive Self-Improvement via Two-Timescale Meta-Skill Evolution to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.05297; inspect its method and evaluation before treating results as production evidence.
medium
README.md
721
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L721
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.05297
[2607.05297] MetaSkill-Evolve: Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution
Recent LLM agents tackle increasingly long-horizon, open-ended tasks, and external skills, reusable procedural knowledge supplied to the agent, further extend this capability. However, a fixed, hand-authored skill is rarely optimal, and cannot adapt to the diversity of tasks an agent encounters. Self-improving agents a...
Zefeng Wang; Minxi Yan; Jinhe Bi; Sikuan Yan; Volker Tresp; Yunpu Ma
2026-07-06
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.05297
2026-08-07T12:31:05
ale-0146
Research Foundations
research-foundations
Paper
📄
SkillOpt-Lite: Better and Faster Agent Self-Evolution via One Line of Vibe
https://arxiv.org/abs/2607.03451
external
arxiv.org
Formalizes agent skill self-evolution as zeroth-order optimization and distills it into a minimal pipeline of file-system trajectory exploration, consensus attribute mining, and independent validation gating, letting a smaller model surpass larger ones on LiveMath and SpreadsheetBench.
Formalizes agent skill self-evolution as zeroth-order optimization and distills it into a minimal pipeline of file-system trajectory exploration, consensus attribute mining, and independent validation gating, letting a smaller model surpass larger ones on LiveMath and SpreadsheetBench.
Formalizes agent skill self-evolution as zeroth-order optimization and distills it into a minimal pipeline of file-system trajectory exploration, consensus attribute mining, and independent validation gating, letting a smaller model surpass larger ones on LiveMath and SpreadsheetBench.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Formalizes agent skill self-evolution as zeroth-order optimization and distills it into a minimal pipeline of file-system trajectory exploration, consensus attribute mining, and independent validation gating, letting a smaller model surpass large...
Use SkillOpt-Lite: Better and Faster Agent Self-Evolution via One Line of Vibe to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.03451; inspect its method and evaluation before treating results as production evidence.
medium
README.md
722
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L722
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.03451
[2607.03451] SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibe
While skill optimization for autonomous agents has gained traction, existing methods rely on complex pipelines. This leaves a fundamental question unaddressed: What constitutes a minimal viable pipeline for skill optimization, where every component is justified by theory or empirical necessity? We formalize skill optim...
Yifei Shen; Bo Li; Xinjie Zhang
2026-07-03
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.03451
2026-08-07T12:31:05
ale-0147
Research Foundations
research-foundations
Paper
📄
Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops
https://arxiv.org/abs/2607.07663
external
arxiv.org
Survey of 1,250 arXiv papers from 2024-2026 organized along two axes, what a self-improvement loop improves and its degree of loop closure, separating bounded evaluable self-refinement from open-ended recursive self-improvement.
Survey of 1,250 arXiv papers from 2024-2026 organized along two axes, what a self-improvement loop improves and its degree of loop closure, separating bounded evaluable self-refinement from open-ended recursive self-improvement.
Survey of 1,250 arXiv papers from 2024-2026 organized along two axes, what a self-improvement loop improves and its degree of loop closure, separating bounded evaluable self-refinement from open-ended recursive self-improvement.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Survey of 1,250 arXiv papers from 2024-2026 organized along two axes, what a self-improvement loop improves and its degree of loop closure, separating bounded evaluable self-refinement from open-ended recursive self-improvement.
Use Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.07663; inspect its method and evaluation before treating results as production evidence.
medium
README.md
723
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L723
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07663
[2607.07663] Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops
AI systems increasingly participate in their own improvement: revising their outputs, adapting their own harnesses during deployment, training on data they generate, and, increasingly, conducting AI research itself. This literature is described under a vocabulary ("self-refine," "self-reward," "self-play," "self-evolve...
Mingguang Chen; Licheng Wang; Bo Qu
2026-07-08
2026
arXiv
arXiv
42 pages, 6 figures
cs.AI
arxiv-api
2607.07663
2026-08-07T12:31:05
ale-0148
Research Foundations
research-foundations
Paper
📄
From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents
https://arxiv.org/abs/2607.07321
external
arxiv.org
EvoSOP has agents distill recurring execution trajectories into reusable standard operating procedures and iteratively optimize the toolset through a construction, merging, evaluation, and pruning lifecycle.
EvoSOP has agents distill recurring execution trajectories into reusable standard operating procedures and iteratively optimize the toolset through a construction, merging, evaluation, and pruning lifecycle.
EvoSOP has agents distill recurring execution trajectories into reusable standard operating procedures and iteratively optimize the toolset through a construction, merging, evaluation, and pruning lifecycle.
Evaluation data is used as the feedback signal for improving loop behavior. EvoSOP has agents distill recurring execution trajectories into reusable standard operating procedures and iteratively optimize the toolset through a construction, merging, evaluation, and pruning lifecycle.
Use From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.07321; inspect its method and evaluation before treating results as production evidence.
medium
README.md
724
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L724
Learn
learn
Understand the field and its boundaries.
workspace;verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07321
[2607.07321] From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents
Tool utilization enables Large Language Model (LLM) agents to interact with the real world and resolve complex tasks. However, existing agent frameworks predominantly rely on static toolsets composed of granular atomic actions (e.g., basic file I/O or single-turn search), which forces agents to reinvent low-level logic...
Haipeng Ding; Yuexiang Xie; Zhewei Wei; Yaliang Li; Bolin Ding
2026-07-08
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.07321
2026-08-07T12:31:05
ale-0149
Research Foundations
research-foundations
Paper
📄
TTHE: Test-Time Harness Evolution
https://arxiv.org/abs/2607.08124
external
arxiv.org
Adapts LLM agents at test time by evolving a population of candidate harnesses (the executable control program around the model) from execution traces, using a label-free agentic proposer and judge to sustain improvements on text-to-SQL and competitive programming while flagging execution-derived proxy reliability as t...
Adapts LLM agents at test time by evolving a population of candidate harnesses (the executable control program around the model) from execution traces, using a label-free agentic proposer and judge to sustain improvements on text-to-SQL and competitive programming while flagging execution-derived proxy reliability as t...
Adapts LLM agents at test time by evolving a population of candidate harnesses (the executable control program around the model) from execution traces, using a label-free agentic proposer and judge to sustain improvements on text-to-SQL and competitive programming while flagging execution-derived proxy reliability as t...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Adapts LLM agents at test time by evolving a population of candidate harnesses (the executable control program around the model) from execution traces, using a label-free agentic proposer and judge to sustain improvements on text-to-SQL and compe...
Use TTHE: Test-Time Harness Evolution to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.08124; inspect its method and evaluation before treating results as production evidence.
medium
README.md
725
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L725
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.08124
[2607.08124] TTHE: Test-Time Harness Evolution
The behavior of an LLM agent is determined not only by the underlying model, but also by its harness: the executable program that constructs context, invokes tools, verifies intermediate results, and recovers from failures. Existing approaches optimize such harnesses before deployment, searching training or development...
Jun Nie; Yonggang Zhang; Jun Song; Qianshu Cai; Dahai Yu; Yike Guo; Xinmei Tian; Bo Han
2026-07-09
2026
arXiv
arXiv
15 pages, 5 figures
cs.SE
arxiv-api
2607.08124
2026-08-07T12:31:05
ale-0150
Research Foundations
research-foundations
Paper
📄
DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment
https://arxiv.org/abs/2607.07820
external
arxiv.org
Introduces DeepSearch-Evolve, where a deep search agent improves by self-distilling its own trajectories inside a deterministic 420K-task verifiable environment, progress verification, grounded reflection, and failure recovery replace teacher trajectories and sparse RL reward, lifting a 9B model to 31.2% BrowseComp and...
Introduces DeepSearch-Evolve, where a deep search agent improves by self-distilling its own trajectories inside a deterministic 420K-task verifiable environment, progress verification, grounded reflection, and failure recovery replace teacher trajectories and sparse RL reward, lifting a 9B model to 31.2% BrowseComp and...
Introduces DeepSearch-Evolve, where a deep search agent improves by self-distilling its own trajectories inside a deterministic 420K-task verifiable environment, progress verification, grounded reflection, and failure recovery replace teacher trajectories and sparse RL reward, lifting a 9B model to 31.2% BrowseComp and...
Verification is promoted from a final check to a loop-control signal. Introduces DeepSearch-Evolve, where a deep search agent improves by self-distilling its own trajectories inside a deterministic 420K-task verifiable environment, progress verification, grounded reflection, and failure recovery replace teacher traject...
Use DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.07820; inspect its method and evaluation before treating results as production evidence.
medium
README.md
726
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L726
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07820
[2607.07820] DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment
Training tool-use agents to improve from their own experience remains challenging, as supervised fine-tuning relies on fixed teacher-distilled trajectories, while sparse-reward reinforcement learning provides weak supervision for long-horizon interactions. We present DeepSearch-Evolve, a self-distillation framework for...
Xinyu Geng; Xuanhua He; Sixiang Chen; Yanjing Xiao; Fan Zhang; Shijue Huang; Haitao Mi; Zhenwen Liang; Tianqing Fang; Yi R. Fung
2026-07-08
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.07820
2026-08-07T12:31:05
ale-0151
Research Foundations
research-foundations
Paper
📄
What Makes a Good Bug Report for an AI Agent?
https://arxiv.org/abs/2607.07593
external
arxiv.org
Statistical analysis of 433 issues plus controlled multi-model experiments showing LLM repair agents succeed more when bug reports carry reproduction scripts, fix suggestions, and fault-localization cues, while longer natural-language reports correlate with lower success, directly informing how a loop's work-discovery ...
Statistical analysis of 433 issues plus controlled multi-model experiments showing LLM repair agents succeed more when bug reports carry reproduction scripts, fix suggestions, and fault-localization cues, while longer natural-language reports correlate with lower success, directly informing how a loop's work-discovery ...
Statistical analysis of 433 issues plus controlled multi-model experiments showing LLM repair agents succeed more when bug reports carry reproduction scripts, fix suggestions, and fault-localization cues, while longer natural-language reports correlate with lower success, directly informing how a loop's work-discovery ...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Statistical analysis of 433 issues plus controlled multi-model experiments showing LLM repair agents succeed more when bug reports carry reproduction scripts, fix suggestions, and fault-localization cues, while longer natural-language reports cor...
Use What Makes a Good Bug Report for an AI Agent? to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.07593; inspect its method and evaluation before treating results as production evidence.
medium
README.md
727
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L727
Learn
learn
Understand the field and its boundaries.
intake
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.07593
[2607.07593] What Makes a Good Bug Report for an AI Agent?
Automated program repair (APR) agents are transitioning from research benchmarks to developer workflows, yet they still begin with bug reports written for human developers. While decades of research have established what makes a good bug report for humans (e.g., steps to reproduce, stack traces), it remains unclear whe...
Lara Khatib; Noble Saji Mathews; Meiyappan Nagappan; Pengyu Nie; Thomas Zimmermann
2026-07-08
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.07593
2026-08-07T12:31:05
ale-0152
Research Foundations
research-foundations
Paper
📄
AutoPersonas: A Multi-Timescale Loop Engine for Open-Ended Persona Evolution
https://arxiv.org/abs/2607.08252
external
arxiv.org
Names self-locking as a runtime failure mode of continuing agent loops, where accumulated state and history pull generation toward stale repetition (over 95% rolling action-repetition across an eight-model 40-day stress test), and proposes a multi-timescale loop that admits divergent material only through evidence-gove...
Names self-locking as a runtime failure mode of continuing agent loops, where accumulated state and history pull generation toward stale repetition (over 95% rolling action-repetition across an eight-model 40-day stress test), and proposes a multi-timescale loop that admits divergent material only through evidence-gove...
Names self-locking as a runtime failure mode of continuing agent loops, where accumulated state and history pull generation toward stale repetition (over 95% rolling action-repetition across an eight-model 40-day stress test), and proposes a multi-timescale loop that admits divergent material only through evidence-gove...
State persistence is explicit enough for repeated runs and handoff. Names self-locking as a runtime failure mode of continuing agent loops, where accumulated state and history pull generation toward stale repetition (over 95% rolling action-repetition across an eight-model 40-day stress test), and proposes a multi-time...
Use AutoPersonas: A Multi-Timescale Loop Engine for Open-Ended Persona Evolution to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.08252; inspect its method and evaluation before treating results as production evidence.
medium
README.md
728
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L728
Learn
learn
Understand the field and its boundaries.
verification;state
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.08252
[2607.08252] AutoPersonas: A Multi-Timescale Loop Engine for Open-Ended Persona Evolution
Long-term persona agents must remain identifiable while adapting to new events, relationships, evidence, and social conditions. We identify self-locking as a runtime failure mode in continuing persona-life loops: locally plausible events keep appearing while the generated life collapses toward familiar environments, we...
Mengchen Li
2026-07-09
2026
arXiv
arXiv
52 pages, 13 figures/tables, ancillary public-safe evaluation artifacts included
cs.AI
arxiv-api
2607.08252
2026-08-07T12:31:05
ale-0153
Research Foundations
research-foundations
Paper
📄
Agentic Data Environments
https://arxiv.org/abs/2607.07397
external
arxiv.org
Vision paper from the IEEE Data Engineering Bulletin reframing data systems as the active execution environment agents operate in, spanning files, APIs, applications, and system state, arguing the substrate under recurring agent loops should both amplify agent capability and enforce safety guarantees that bound the cos...
Vision paper from the IEEE Data Engineering Bulletin reframing data systems as the active execution environment agents operate in, spanning files, APIs, applications, and system state, arguing the substrate under recurring agent loops should both amplify agent capability and enforce safety guarantees that bound the cos...
Vision paper from the IEEE Data Engineering Bulletin reframing data systems as the active execution environment agents operate in, spanning files, APIs, applications, and system state, arguing the substrate under recurring agent loops should both amplify agent capability and enforce safety guarantees that bound the cos...
State persistence is explicit enough for repeated runs and handoff. Vision paper from the IEEE Data Engineering Bulletin reframing data systems as the active execution environment agents operate in, spanning files, APIs, applications, and system state, arguing the substrate under recurring agent loops should both ampli...
Use Agentic Data Environments to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.07397; inspect its method and evaluation before treating results as production evidence.
medium
README.md
729
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L729
Learn
learn
Understand the field and its boundaries.
state;budget
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
http://sites.computer.org/debull/A26mar/A26MAR-CD.pdf#page=7
[2607.07397] Agentic Data Environments
Autonomous agents promise substantial gains in speed, scale, and labor efficiency, but their failures can impose abrupt and often irreversible costs. The central challenge for agentic automation is therefore to increase the benefits of automation while bounding the consequences of failure. While databases remain centra...
Elaine Ang; Chenxi Huang; Georgios Liargkovas; Jerry Liu; Jinhui Liu; Nikos Pagonas; Charlie Summers; Haonan Wang; Jiakai Xu; Tianle Zhou; Yusen Zhang; Zhou Yu; Zhuo Zhang; Tianyi Peng; Kostis Kaffes; Eugene Wu
2026-03
2026
IEEE Data Engineering Bulletin 50(1)
IEEE
Published in IEEE Data Engineering Bulletin 50(1); the linked arXiv record remains available for open access.
cs.AI
IEEE Data Engineering Bulletin record
2607.07397
2026-08-07T12:31:05
ale-0154
Research Foundations
research-foundations
Paper
📄
Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptation
https://arxiv.org/abs/2607.08938
external
arxiv.org
Meta agent maps observed failure modes to harness adaptation strategies, letting small-model agents recover ~90% of frontier-LLM performance at ~4% of the cost, the harness itself becomes the optimization target.
Meta agent maps observed failure modes to harness adaptation strategies, letting small-model agents recover ~90% of frontier-LLM performance at ~4% of the cost, the harness itself becomes the optimization target.
Meta agent maps observed failure modes to harness adaptation strategies, letting small-model agents recover ~90% of frontier-LLM performance at ~4% of the cost, the harness itself becomes the optimization target.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Meta agent maps observed failure modes to harness adaptation strategies, letting small-model agents recover ~90% of frontier-LLM performance at ~4% of the cost, the harness itself becomes the optimization target.
Use Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptation to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.08938; inspect its method and evaluation before treating results as production evidence.
medium
README.md
730
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L730
Learn
learn
Understand the field and its boundaries.
budget
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.08938
[2607.08938] Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptation
Frontier LLM agents are automating many business tasks, but their high inference cost makes large-scale deployment unsustainable. Small language models (SLMs) offer a cheaper alternative, yet they typically fall short when swapped into a harness designed for a frontier LLM. We show that for many routine business tasks,...
Chenyang Yang; Xinran Zhao; Tongshuang Wu; Christian Kästner
2026-07-09
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.08938
2026-08-07T12:31:05
ale-0155
Research Foundations
research-foundations
Paper
📄
Inside the Skill Market: From Software Engineering Activities to Reusable Agent Skills
https://arxiv.org/abs/2607.09065
external
arxiv.org
First large-scale empirical study of public agent-skill repositories and marketplaces, characterizing which software-engineering activities get packaged as reusable skills, their coverage across the development lifecycle, how they evolve, and how they are evaluated - an activity-centric map of the skills layer that age...
First large-scale empirical study of public agent-skill repositories and marketplaces, characterizing which software-engineering activities get packaged as reusable skills, their coverage across the development lifecycle, how they evolve, and how they are evaluated - an activity-centric map of the skills layer that age...
First large-scale empirical study of public agent-skill repositories and marketplaces, characterizing which software-engineering activities get packaged as reusable skills, their coverage across the development lifecycle, how they evolve, and how they are evaluated - an activity-centric map of the skills layer that age...
Connects Loop Engineering to prior agent-loop and feedback-loop research. First large-scale empirical study of public agent-skill repositories and marketplaces, characterizing which software-engineering activities get packaged as reusable skills, their coverage across the development lifecycle, how they evolve, and how...
Use Inside the Skill Market: From Software Engineering Activities to Reusable Agent Skills to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.09065; inspect its method and evaluation before treating results as production evidence.
medium
README.md
731
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L731
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.09065
[2607.09065] Inside the Skill Market: From Software Engineering Activities to Reusable Agent Skills
Software engineering (abbrev. SE) has continuously evolved through increasingly powerful forms of reuse, from source code and libraries to components and services. Recent advances in AI agents have introduced a potentially new reusable artifact: skills. Emerging agent skill repositories and marketplaces enable develope...
Jialun Cao; Xinru Yan; Songqiang Chen; Yaojie Lu; Zhongxin Liu; Shing-Chi Cheung
2026-07-10
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.09065
2026-08-07T12:31:05
ale-0156
Research Foundations
research-foundations
Blog
📝
Harness Engineering for Self-Improvement
https://lilianweng.github.io/posts/2026-07-04-harness/
external
lilianweng.github.io
Lilian Weng's deep-dive arguing the harness, the system surrounding a base model that orchestrates execution, matters as much as raw intelligence for recursive self-improvement, with a taxonomy of harness components and failure modes.
Lilian Weng's deep-dive arguing the harness, the system surrounding a base model that orchestrates execution, matters as much as raw intelligence for recursive self-improvement, with a taxonomy of harness components and failure modes.
Lilian Weng's deep-dive arguing the harness, the system surrounding a base model that orchestrates execution, matters as much as raw intelligence for recursive self-improvement, with a taxonomy of harness components and failure modes.
Orchestration and control flow are made explicit and inspectable. Lilian Weng's deep-dive arguing the harness, the system surrounding a base model that orchestrates execution, matters as much as raw intelligence for recursive self-improvement, with a taxonomy of harness components and failure modes.
Use Harness Engineering for Self-Improvement to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Contextual source from lilianweng.github.io; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
732
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L732
Learn
learn
Understand the field and its boundaries.
delegation
builder
cross-layer
enabling
practitioner-analysis
B
ok
https://lilianweng.github.io/posts/2026-07-04-harness/
Harness Engineering for Self-Improvement | Lil'Log
The concept of recursive self-improvement (RSI) dates back to I. J. Good (1965), where he defined an “ultraintelligent machine” as a system that can surpass humans in all intellectual activities and design better machines to improve itself. Yudkowsky (2008) used the phrase “recursive self-improvement” for a specific fe...
Lilian Weng
2026-07-04
2026
lilianweng.github.io
html-meta
2026-08-07T12:31:05
ale-0157
Research Foundations
research-foundations
Paper
📄
Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime
https://arxiv.org/abs/2607.11346
external
arxiv.org
Compiles safety-critical standard operating procedures into executable pseudo-code run by a program-guided stack machine that pages the active frame while the LLM does semantic execution, finding runtime guidance is capability-gated (it helps strong models and harms weak ones) across a six-model, seven-domain SOPBench ...
Compiles safety-critical standard operating procedures into executable pseudo-code run by a program-guided stack machine that pages the active frame while the LLM does semantic execution, finding runtime guidance is capability-gated (it helps strong models and harms weak ones) across a six-model, seven-domain SOPBench ...
Compiles safety-critical standard operating procedures into executable pseudo-code run by a program-guided stack machine that pages the active frame while the LLM does semantic execution, finding runtime guidance is capability-gated (it helps strong models and harms weak ones) across a six-model, seven-domain SOPBench ...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Compiles safety-critical standard operating procedures into executable pseudo-code run by a program-guided stack machine that pages the active frame while the LLM does semantic execution, finding runtime guidance is capability-gated (it helps str...
Use Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.11346; inspect its method and evaluation before treating results as production evidence.
medium
README.md
733
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L733
2026-07-15
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.11346
[2607.11346] Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime for Procedural LLM Agents
Enterprise agents must follow long-horizon, conditional, safety-critical standard operating procedures (SOPs). We compile machine-readable SOP constraints into executable pseudo-code and run them with a program-guided (PG) stack machine that pages the active frame while an LLM performs semantic execution. A three-arm S...
Chenglin Yu; Li Yin; Qingxin Fan; Ying Yu; RunyangRay Zhong; Ming Li
2026-07-13
2026
arXiv
arXiv
9 pages, 3 figures, 5 tables
cs.AI
arxiv-api
2607.11346
2026-08-07T12:31:05
ale-0158
Research Foundations
research-foundations
Paper
📄
Mako: A Self-Evolving Agentic Operating System for Autonomous Web Exploitation
https://arxiv.org/abs/2607.11288
external
arxiv.org
Industrial self-evolving agentic OS that treats exploit capability as a mutable, versioned kernel: the agent observes its own failures, synthesizes new capabilities, proves them against a live target, and hot-loads them back, a self-improvement loop with in-loop verification.
Industrial self-evolving agentic OS that treats exploit capability as a mutable, versioned kernel: the agent observes its own failures, synthesizes new capabilities, proves them against a live target, and hot-loads them back, a self-improvement loop with in-loop verification.
Industrial self-evolving agentic OS that treats exploit capability as a mutable, versioned kernel: the agent observes its own failures, synthesizes new capabilities, proves them against a live target, and hot-loads them back, a self-improvement loop with in-loop verification.
Verification is promoted from a final check to a loop-control signal. Industrial self-evolving agentic OS that treats exploit capability as a mutable, versioned kernel: the agent observes its own failures, synthesizes new capabilities, proves them against a live target, and hot-loads them back, a self-improvement loop ...
Use Mako: A Self-Evolving Agentic Operating System for Autonomous Web Exploitation to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.11288; inspect its method and evaluation before treating results as production evidence.
medium
README.md
734
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L734
2026-07-15
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.11288
[2607.11288] Mako: A Self-Evolving Agentic Operating System (SE-AOS) for Autonomous Web Exploitation
We introduce the Self-Evolving Agentic Operating System (SE-AOS): a new class of AI agent that treats exploit capability as a mutable, versioned kernel it extends at runtime, observing its own failures, synthesising new capabilities, proving them against a live target, and hot-loading them back into itself. Mako is the...
Praneeth Narisetty; Shiva Nagendra Babu Kore
2026-07-13
2026
arXiv
arXiv
13 pages, 10 figures, 8 tables
cs.CR
arxiv-api
2607.11288
2026-08-07T12:31:05
ale-0159
Research Foundations
research-foundations
Paper
📄
How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study
https://arxiv.org/abs/2607.10856
external
arxiv.org
Mixed-methods study of how practitioners actually design software-engineering agents, surfacing the recurring loop, harness, and verification decisions teams make and where their mental models diverge from benchmark assumptions.
Mixed-methods study of how practitioners actually design software-engineering agents, surfacing the recurring loop, harness, and verification decisions teams make and where their mental models diverge from benchmark assumptions.
Mixed-methods study of how practitioners actually design software-engineering agents, surfacing the recurring loop, harness, and verification decisions teams make and where their mental models diverge from benchmark assumptions.
Verification is promoted from a final check to a loop-control signal. Mixed-methods study of how practitioners actually design software-engineering agents, surfacing the recurring loop, harness, and verification decisions teams make and where their mental models diverge from benchmark assumptions.
Use How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.10856; inspect its method and evaluation before treating results as production evidence.
medium
README.md
735
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L735
2026-07-15
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.10856
[2607.10856] How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study
The rise of Software Engineering (SE) agents, i.e., LLM-based agents that can understand large codebases and carry out engineering tasks with limited human intervention, has been marked by rapid advances and adoption, but little is known about how developers build these systems in practice: existing studies mine reposi...
Yunbo Lyu; David Williams; Jieke Shi; Zhensu Sun; Chao Peng; Zhou Yang; Federica Sarro; David Lo
2026-07-12
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.10856
2026-08-07T12:31:05
ale-0160
Research Foundations
research-foundations
Paper
📄
Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries
https://arxiv.org/abs/2607.10113
external
arxiv.org
Survey and taxonomy of how agent skill libraries are created, evaluated, retired, and reused over time, organizing the fast-growing self-evolving-skills literature into a lifecycle framework.
Survey and taxonomy of how agent skill libraries are created, evaluated, retired, and reused over time, organizing the fast-growing self-evolving-skills literature into a lifecycle framework.
Survey and taxonomy of how agent skill libraries are created, evaluated, retired, and reused over time, organizing the fast-growing self-evolving-skills literature into a lifecycle framework.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Survey and taxonomy of how agent skill libraries are created, evaluated, retired, and reused over time, organizing the fast-growing self-evolving-skills literature into a lifecycle framework.
Use Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.10113; inspect its method and evaluation before treating results as production evidence.
medium
README.md
736
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L736
2026-07-15
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-paper
A
ok
https://openreview.net/forum?id=cjU3YbcRr8
[2607.10113] Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries
Large language model agents increasingly store reusable procedures outside the model. These reusable procedures are often called \emph{skills}: they may be code functions, natural-language instructions, SKILL.md packages, workflow graphs, or learned adapters that a future agent can retrieve and invoke. This taxonomy-dr...
Yubo Li
2026
2026
Transactions on Machine Learning Research (TMLR)
OpenReview
Accepted at Transactions on Machine Learning Research (TMLR); the linked arXiv record is the available paper version.
cs.AI
Current arXiv acceptance note and OpenReview record
2607.10113
2026-08-07T12:31:05
ale-0161
Research Foundations
research-foundations
Paper
📄
SIA: Self Improving AI with Harness & Weight Updates
https://arxiv.org/abs/2605.27276
external
arxiv.org
Co-evolves a task agent's harness and model weights through a meta-agent, target agent, and feedback agent that evaluate outcomes and carry improvements across generations.
Co-evolves a task agent's harness and model weights through a meta-agent, target agent, and feedback agent that evaluate outcomes and carry improvements across generations.
Co-evolves a task agent's harness and model weights through a meta-agent, target agent, and feedback agent that evaluate outcomes and carry improvements across generations.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Co-evolves a task agent's harness and model weights through a meta-agent, target agent, and feedback agent that evaluate outcomes and carry improvements across generations.
Use SIA: Self Improving AI with Harness & Weight Updates to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2605.27276; inspect its method and evaluation before treating results as production evidence.
medium
README.md
740
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L740
2026-07-18
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2605.27276
[2605.27276] SIA: Self Improving AI with Harness & Weight Updates
Humans are the bottleneck in building and improving AI. Both the models and the agents that wrap them are written, tuned, and corrected by people. The long-horizon goal of an AI that can figure out how to improve itself remains open. Two largely disjoint research lines attack this bottleneck. The harness-update school ...
Prannay Hebbar; Yogendra Manawat; Samuel Verboomen; Alesia Ivanova; Selvam Palanimalai; Kunal Bhatia; Vignesh Baskaran
2026-05-26
2026
arXiv
arXiv
cs.AI
arxiv-api
2605.27276
2026-08-07T12:31:05
ale-0162
Research Foundations
research-foundations
Paper
📄
Self-Improvements in Modern Agentic Systems: A Survey
https://arxiv.org/abs/2607.13104
external
arxiv.org
Survey from the Zhuge/Schmidhuber group that frames a modern agent as a foundation model coupled to an operational scaffold (prompts, memory, tools, control logic) and organizes self-improvement research by which scaffold component gets updated and what signal drives the update.
Survey from the Zhuge/Schmidhuber group that frames a modern agent as a foundation model coupled to an operational scaffold (prompts, memory, tools, control logic) and organizes self-improvement research by which scaffold component gets updated and what signal drives the update.
Survey from the Zhuge/Schmidhuber group that frames a modern agent as a foundation model coupled to an operational scaffold (prompts, memory, tools, control logic) and organizes self-improvement research by which scaffold component gets updated and what signal drives the update.
Persistent memory is treated as an external runtime artifact. Survey from the Zhuge/Schmidhuber group that frames a modern agent as a foundation model coupled to an operational scaffold (prompts, memory, tools, control logic) and organizes self-improvement research by which scaffold component gets updated and what sign...
Use Self-Improvements in Modern Agentic Systems: A Survey to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.13104; inspect its method and evaluation before treating results as production evidence.
medium
README.md
741
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L741
2026-07-22
Learn
learn
Understand the field and its boundaries.
workspace;context
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.13104
[2607.13104] Self-Improvements in Modern Agentic Systems: A Survey
Self-improving autonomous agents are moving from research prototypes to deployed systems. The primary goal is controllable evolution, or adaptation, from experience with minimal or even no human input. This survey frames modern self-improving agents as adaptive systems that convert experience into accumulated capabilit...
Zhe Ren; Yimeng Chen; Dandan Guo; Guowei Rong; Tonghui Li; R. B. Xiong; Qingfeng Lan; Wenyi Wang; Li Nanbo; Yibo Yang; Mingchen Zhuge; Jürgen Schmidhuber
2026-07-14
2026
arXiv
arXiv
97 pages, 12 figures. Project page: https://selfimproving-agent.github.io/ Repository: https://github.com/selfimproving-agent/awesome-Self-Improving-Agents
cs.AI
arxiv-api
2607.13104
2026-08-07T12:31:05
ale-0163
Research Foundations
research-foundations
Paper
📄
Self-Evolving Agent Harnesses via Gated Semantic Quality-Diversity
https://arxiv.org/abs/2607.13683
external
arxiv.org
Evolves the agent harness (prompts, injected knowledge, runtime config) with model weights frozen: an LLM diagnoses failures and proposes patches while deterministic code owns all sampling and significance testing, and accepted patches populate a pathology-keyed quality-diversity archive, with sealed-test gains of 9-15...
Evolves the agent harness (prompts, injected knowledge, runtime config) with model weights frozen: an LLM diagnoses failures and proposes patches while deterministic code owns all sampling and significance testing, and accepted patches populate a pathology-keyed quality-diversity archive, with sealed-test gains of 9-15...
Evolves the agent harness (prompts, injected knowledge, runtime config) with model weights frozen: an LLM diagnoses failures and proposes patches while deterministic code owns all sampling and significance testing, and accepted patches populate a pathology-keyed quality-diversity archive, with sealed-test gains of 9-15...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Evolves the agent harness (prompts, injected knowledge, runtime config) with model weights frozen: an LLM diagnoses failures and proposes patches while deterministic code owns all sampling and significance testing, and accepted patches populate a...
Use Self-Evolving Agent Harnesses via Gated Semantic Quality-Diversity to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.13683; inspect its method and evaluation before treating results as production evidence.
medium
README.md
742
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L742
2026-07-22
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.13683
[2607.13683] HarnessBank: Semantic Gene-Bank Search with Gated Verification for Agent-Harness Self-Evolution
Large Language Models (LLMs) have enabled capable agents across diverse applications. Beyond the foundation model, the performance of an agent is governed by the surrounding agent harness, including prompts, tools, control loops, etc. Automatically evolving this harness offers a promising pathway to agent improvement, ...
Xiaotian Luo; Dizhan Xue; Fengxingyu Wang; Chuanrui Hu; Yafeng Deng
2026-07-15
2026
arXiv
arXiv
9 pages, 4 figures, 3 tables
cs.CL
arxiv-api
2607.13683
2026-08-07T12:31:05
ale-0164
Research Foundations
research-foundations
Paper
📄
XScientist: A Git-Like Research Protocol for Long-Running Autonomous Scientific Discovery
https://arxiv.org/abs/2607.12301
external
arxiv.org
Treats long-running autonomous research as a continuous observable pipeline: exploration recorded as portable checkpointed artifacts with code, outputs, and evidence links, plus repair loops and human-in-the-loop quality gates, so failed branches stay inspectable. Single-author work, content is on-target for long-horiz...
Treats long-running autonomous research as a continuous observable pipeline: exploration recorded as portable checkpointed artifacts with code, outputs, and evidence links, plus repair loops and human-in-the-loop quality gates, so failed branches stay inspectable. Single-author work, content is on-target for long-horiz...
Treats long-running autonomous research as a continuous observable pipeline: exploration recorded as portable checkpointed artifacts with code, outputs, and evidence links, plus repair loops and human-in-the-loop quality gates, so failed branches stay inspectable. Single-author work, content is on-target for long-horiz...
Checkpointed state makes long-running agent work recoverable across failures. Treats long-running autonomous research as a continuous observable pipeline: exploration recorded as portable checkpointed artifacts with code, outputs, and evidence links, plus repair loops and human-in-the-loop quality gates, so failed bran...
Use XScientist: A Git-Like Research Protocol for Long-Running Autonomous Scientific Discovery to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.12301; inspect its method and evaluation before treating results as production evidence.
medium
README.md
743
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L743
2026-07-22
Learn
learn
Understand the field and its boundaries.
intake;state;escalation
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.12301
[2607.12301] XScientist: A Git-Like Research Protocol for Long-Running Autonomous Scientific Discovery
Autonomous research systems are often evaluated as one-shot paper generators: given a topic, they produce a manuscript and a small set of experiment logs. This framing hides the operational problem that makes such systems difficult to trust: research is long-running, branching, failure-prone, and dependent on auditable...
Jixiang Luo
2026-07-14
2026
arXiv
arXiv
cs.SE
arxiv-api
2607.12301
2026-08-07T12:31:05
ale-0165
Research Foundations
research-foundations
Paper
📄
Knowledge-Centric Self-Improvement
https://arxiv.org/abs/2607.19592
external
arxiv.org
Proposes a knowledge-centric alternative to agent-centric self-improvement: agents stay generic and disposable while the persistent, improving object is a curated shared knowledge base that agents write evidence-grounded insights into and then distill, reporting higher solve rates at lower cost with knowledge that tran...
Proposes a knowledge-centric alternative to agent-centric self-improvement: agents stay generic and disposable while the persistent, improving object is a curated shared knowledge base that agents write evidence-grounded insights into and then distill, reporting higher solve rates at lower cost with knowledge that tran...
Proposes a knowledge-centric alternative to agent-centric self-improvement: agents stay generic and disposable while the persistent, improving object is a curated shared knowledge base that agents write evidence-grounded insights into and then distill, reporting higher solve rates at lower cost with knowledge that tran...
State persistence is explicit enough for repeated runs and handoff. Proposes a knowledge-centric alternative to agent-centric self-improvement: agents stay generic and disposable while the persistent, improving object is a curated shared knowledge base that agents write evidence-grounded insights into and then distill,...
Use Knowledge-Centric Self-Improvement to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.19592; inspect its method and evaluation before treating results as production evidence.
medium
README.md
744
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L744
2026-07-23
Learn
learn
Understand the field and its boundaries.
state;budget
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.19592
[2607.19592] Knowledge-Centric Self-Improvement
Self-improving AI systems typically treat the agent as the object that improves, by optimizing prompts, workflows, harnesses, or even the agent's own code. This agent-centric view can make improvements expensive to maintain and difficult to transfer, because gains become tied to a particular agent design, task distribu...
Xuefei Julie Wang; Lauren Hyoseo Yoon; Chengrui Qu; Amanda Zichang Wang; Atharva Sehgal; Eric Mazumdar; Yisong Yue
2026-07-21
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.19592
2026-08-07T12:31:05
ale-0166
Research Foundations
research-foundations
Paper
📄
OpenForgeRL: Train Harness-native Agents in Any Environment
https://arxiv.org/abs/2607.21557
external
arxiv.org
Open-source framework for end-to-end SFT/RL training of agents that run inside real inference harnesses (Claude Code, OpenClaw) via a lightweight proxy that serves the harness's model calls while recording them as training data, with Kubernetes-orchestrated distributed rollouts. Directly attacks the gap between harness...
Open-source framework for end-to-end SFT/RL training of agents that run inside real inference harnesses (Claude Code, OpenClaw) via a lightweight proxy that serves the harness's model calls while recording them as training data, with Kubernetes-orchestrated distributed rollouts. Directly attacks the gap between harness...
Open-source framework for end-to-end SFT/RL training of agents that run inside real inference harnesses (Claude Code, OpenClaw) via a lightweight proxy that serves the harness's model calls while recording them as training data, with Kubernetes-orchestrated distributed rollouts. Directly attacks the gap between harness...
Orchestration and control flow are made explicit and inspectable. Open-source framework for end-to-end SFT/RL training of agents that run inside real inference harnesses (Claude Code, OpenClaw) via a lightweight proxy that serves the harness's model calls while recording them as training data, with Kubernetes-orchestra...
Use OpenForgeRL: Train Harness-native Agents in Any Environment to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.21557; inspect its method and evaluation before treating results as production evidence.
medium
README.md
745
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L745
2026-07-24
Learn
learn
Understand the field and its boundaries.
delegation;state
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21557
[2607.21557] OpenForgeRL: Train Harness-native Agents in Any Environment
Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks cannot natively express...
Xiao Yu; Baolin Peng; Ruize Xu; Hao Zou; Qianhui Wu; Hao Cheng; Wenlin Yao; Nikhil Singh; Zhou Yu; Jianfeng Gao
2026-07-23
2026
arXiv
arXiv
updated the paper header to show ICLR2027 instead of ICLR2026 (already past)
cs.AI
arxiv-api
2607.21557
2026-08-07T12:31:05
ale-0167
Research Foundations
research-foundations
Paper
📄
AREX: Towards a Recursively Self-Improving Agent for Deep Research
https://arxiv.org/abs/2607.21461
external
arxiv.org
Family of recursively self-improving deep-research agents built on the discovery-verification asymmetry: an inner research loop gathers evidence while an outer loop audits the answer constraint-wise, flags unresolved claims, and launches targeted follow-up research, with autonomously learned compression of interaction ...
Family of recursively self-improving deep-research agents built on the discovery-verification asymmetry: an inner research loop gathers evidence while an outer loop audits the answer constraint-wise, flags unresolved claims, and launches targeted follow-up research, with autonomously learned compression of interaction ...
Family of recursively self-improving deep-research agents built on the discovery-verification asymmetry: an inner research loop gathers evidence while an outer loop audits the answer constraint-wise, flags unresolved claims, and launches targeted follow-up research, with autonomously learned compression of interaction ...
Verification is promoted from a final check to a loop-control signal. Family of recursively self-improving deep-research agents built on the discovery-verification asymmetry: an inner research loop gathers evidence while an outer loop audits the answer constraint-wise, flags unresolved claims, and launches targeted fol...
Use AREX: Towards a Recursively Self-Improving Agent for Deep Research to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.21461; inspect its method and evaluation before treating results as production evidence.
medium
README.md
746
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L746
2026-07-24
Learn
learn
Understand the field and its boundaries.
intake;verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21461
[2607.21461] AREX: Towards a Recursively Self-Improving Agent for Deep Research
Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do more than simply searc...
Shuqi Lu; Chaofan Li; Kun Luo; Zhang Zhang; Hui Wang; Hongwang Xiao; Lei Xiong; Jiahao Wang; Sen Wang; Xiyan Jiang; Wanli Li; Yuyang Hu; Hongjin Qian; Bingyu Yan; Jianlyu Chen; Ziyi Xia; Yingxia Shao; Kang Liu; Zhicheng Dou; Di He; Chaozhuo Li; Qiwei Ye; Zhongyuan Wang; Zheng Liu
2026-07-23
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.21461
2026-08-07T12:31:05
ale-0168
Research Foundations
research-foundations
Paper
📄
Workflow-Localized Mechanism Learning: Attribution-Guided Repair and Knowledge Reuse for Structured Agent Skills
https://arxiv.org/abs/2607.20999
external
arxiv.org
Optimizes Agent Skills (reusable procedural knowledge for frozen-model agents) by jointly resolving where a workflow failed, which mechanism caused it, and which third-party Skill knowledge to reuse locally: node-mechanism attribution pinpoints the failed node and smallest valid edit target, bounded patches apply the r...
Optimizes Agent Skills (reusable procedural knowledge for frozen-model agents) by jointly resolving where a workflow failed, which mechanism caused it, and which third-party Skill knowledge to reuse locally: node-mechanism attribution pinpoints the failed node and smallest valid edit target, bounded patches apply the r...
Optimizes Agent Skills (reusable procedural knowledge for frozen-model agents) by jointly resolving where a workflow failed, which mechanism caused it, and which third-party Skill knowledge to reuse locally: node-mechanism attribution pinpoints the failed node and smallest valid edit target, bounded patches apply the r...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Optimizes Agent Skills (reusable procedural knowledge for frozen-model agents) by jointly resolving where a workflow failed, which mechanism caused it, and which third-party Skill knowledge to reuse locally: node-mechanism attribution pinpoints t...
Use Workflow-Localized Mechanism Learning: Attribution-Guided Repair and Knowledge Reuse for Structured Agent Skills to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.20999; inspect its method and evaluation before treating results as production evidence.
medium
README.md
747
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L747
2026-07-24
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.20999
[2607.20999] Workflow-Localized Mechanism Learning: Attribution-Guided Repair and Knowledge Reuse for Structured Agent Skills
Agent Skills package reusable procedural knowledge as external artifacts for frozen language-model agents, yet existing optimizers do not jointly resolve where a failure occurs in a workflow, which mechanism caused it, and how relevant knowledge from third-party Skills should be reused locally. We introduce Workflow-Lo...
Zibin Lin; Shengli Zhang; Taotao Wang; Yihan Xia; Deen Ma; Guofu Liao
2026-07-23
2026
arXiv
arXiv
8 pages, 3 figures
cs.AI
arxiv-api
2607.20999
2026-08-07T12:31:05
ale-0169
Research Foundations
research-foundations
Paper
📄
Sample-Efficient Learning from Agent Experience
https://arxiv.org/abs/2607.21051
external
arxiv.org
Uses experience distillation to internalize an agent's own interaction histories into model weights, so in-context experience gains persist after the experience leaves the context window, matching RL-style improvement with far fewer environment interactions.
Uses experience distillation to internalize an agent's own interaction histories into model weights, so in-context experience gains persist after the experience leaves the context window, matching RL-style improvement with far fewer environment interactions.
Uses experience distillation to internalize an agent's own interaction histories into model weights, so in-context experience gains persist after the experience leaves the context window, matching RL-style improvement with far fewer environment interactions.
Context is managed as durable loop state rather than a single prompt payload. Uses experience distillation to internalize an agent's own interaction histories into model weights, so in-context experience gains persist after the experience leaves the context window, matching RL-style improvement with far fewer environme...
Use Sample-Efficient Learning from Agent Experience to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.21051; inspect its method and evaluation before treating results as production evidence.
medium
README.md
748
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L748
2026-07-24
Learn
learn
Understand the field and its boundaries.
context;state
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21051
[2607.21051] Sample-Efficient Learning from Agent Experience
Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once that experience is re...
Chenhui Gou; Haoqin Tu; Yunhao Fang; Jianfei Cai; Hamid Rezatofighi
2026-07-23
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.21051
2026-08-07T12:31:05
ale-0170
Research Foundations
research-foundations
Paper
📄
PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning
https://arxiv.org/abs/2607.21419
external
arxiv.org
Policy-centric training paradigm that reframes skills as a dynamic training-time scaffold for long-horizon agent RL: rollout groups from the latest policy become evidence cards, task-specific evaluation adjusts the context for subsequent rollouts, and guidance is pruned as the policy strengthens before the scaffold is ...
Policy-centric training paradigm that reframes skills as a dynamic training-time scaffold for long-horizon agent RL: rollout groups from the latest policy become evidence cards, task-specific evaluation adjusts the context for subsequent rollouts, and guidance is pruned as the policy strengthens before the scaffold is ...
Policy-centric training paradigm that reframes skills as a dynamic training-time scaffold for long-horizon agent RL: rollout groups from the latest policy become evidence cards, task-specific evaluation adjusts the context for subsequent rollouts, and guidance is pruned as the policy strengthens before the scaffold is ...
Evaluation data is used as the feedback signal for improving loop behavior. Policy-centric training paradigm that reframes skills as a dynamic training-time scaffold for long-horizon agent RL: rollout groups from the latest policy become evidence cards, task-specific evaluation adjusts the context for subsequent rollou...
Use PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.21419; inspect its method and evaluation before treating results as production evidence.
medium
README.md
749
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L749
2026-07-24
Learn
learn
Understand the field and its boundaries.
context;verification;budget
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21419
[2607.21419] PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning
In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization. Existing skill-centric methods improve exploration by optimizing, filtering, or internalizing reusable skills. However, they remain cent...
Yipeng Shi; Zhipeng Ma; Yue Wang; Qitai Tan; Yang Li; Peng Chen; Zhengzhou Zhu
2026-07-23
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.21419
2026-08-07T12:31:05
ale-0171
Research Foundations
research-foundations
Paper
📄
From Agent Failures to Text Policies: What Works and What Breaks
https://arxiv.org/abs/2607.20668
external
arxiv.org
Applies TextGrad-style natural-language policy learning to agent failure trajectories and separates two abilities usually conflated: following a useful text policy versus learning one from experience. Finds a clear gap, human-written policies lift frozen 7B agents by about 5 points, while policies auto-learned from the...
Applies TextGrad-style natural-language policy learning to agent failure trajectories and separates two abilities usually conflated: following a useful text policy versus learning one from experience. Finds a clear gap, human-written policies lift frozen 7B agents by about 5 points, while policies auto-learned from the...
Applies TextGrad-style natural-language policy learning to agent failure trajectories and separates two abilities usually conflated: following a useful text policy versus learning one from experience. Finds a clear gap, human-written policies lift frozen 7B agents by about 5 points, while policies auto-learned from the...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Applies TextGrad-style natural-language policy learning to agent failure trajectories and separates two abilities usually conflated: following a useful text policy versus learning one from experience. Finds a clear gap, human-written policies lif...
Use From Agent Failures to Text Policies: What Works and What Breaks to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.20668; inspect its method and evaluation before treating results as production evidence.
medium
README.md
750
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L750
2026-07-25
Learn
learn
Understand the field and its boundaries.
escalation
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.20668
[2607.20668] From Agent Failures to Text Policies: What Works and What Breaks
TextGrad improves language-model systems by revising text from feedback. Its core thesis is that natural-language feedback can act as a gradient for optimizing text components without changing model weights. Applying it to agents is harder because feedback arrives only after a sequence of actions, making it difficult t...
Jaideep Ray; Ankit Goyal
2026-07-22
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.20668
2026-08-07T12:31:05
ale-0172
Research Foundations
research-foundations
Paper
📄
The Regression Tax: Decomposing Why Skills Help and Hurt LLM Agents
https://arxiv.org/abs/2607.22520
external
arxiv.org
Nearly 6,000 runs across two office-automation benchmarks and three model harness stacks measuring what average success rate hides: skills also cause regressions (tasks solved without the skill, failed with it). Finding that should change how people ship Skills -- the best-performing skills win primarily by regressing ...
Nearly 6,000 runs across two office-automation benchmarks and three model harness stacks measuring what average success rate hides: skills also cause regressions (tasks solved without the skill, failed with it). Finding that should change how people ship Skills -- the best-performing skills win primarily by regressing ...
Nearly 6,000 runs across two office-automation benchmarks and three model harness stacks measuring what average success rate hides: skills also cause regressions (tasks solved without the skill, failed with it). Finding that should change how people ship Skills -- the best-performing skills win primarily by regressing ...
The work turns loop quality into a measurable task or score. Nearly 6,000 runs across two office-automation benchmarks and three model harness stacks measuring what average success rate hides: skills also cause regressions (tasks solved without the skill, failed with it). Finding that should change how people ship Skil...
Use The Regression Tax: Decomposing Why Skills Help and Hurt LLM Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.22520; inspect its method and evaluation before treating results as production evidence.
medium
README.md
751
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L751
2026-07-28
Learn
learn
Understand the field and its boundaries.
context;verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22520
[2607.22520] The Regression Tax: Decomposing Why Skills Help and Hurt LLM Agents
Adding procedural skills to an LLM agent is typically evaluated by average improvement in task success. However, this metric hides an important cost: skills can also make agents worse. We measure both sides by comparing agents with and without skills across nearly 6,000 runs spanning two office automation benchmarks an...
Darshan Tank; Baran Nama
2026-07-24
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.22520
2026-08-07T12:31:05
ale-0173
Research Foundations
research-foundations
Paper
📄
Learning on the Job: Continual Learning from Deployment Feedback for Frozen-Weights Agents
https://arxiv.org/abs/2607.22157
external
arxiv.org
Shows ordinary production feedback -- one-bit outcome verdicts and after-the-fact corrections -- is a sufficient signal for continual learning when a frozen model is paired with external memory distilling each episode into retrievable natural-language rules. On tau-bench banking, against a static-RAG control over the f...
Shows ordinary production feedback -- one-bit outcome verdicts and after-the-fact corrections -- is a sufficient signal for continual learning when a frozen model is paired with external memory distilling each episode into retrievable natural-language rules. On tau-bench banking, against a static-RAG control over the f...
Shows ordinary production feedback -- one-bit outcome verdicts and after-the-fact corrections -- is a sufficient signal for continual learning when a frozen model is paired with external memory distilling each episode into retrievable natural-language rules. On tau-bench banking, against a static-RAG control over the f...
Persistent memory is treated as an external runtime artifact. Shows ordinary production feedback -- one-bit outcome verdicts and after-the-fact corrections -- is a sufficient signal for continual learning when a frozen model is paired with external memory distilling each episode into retrievable natural-language rules....
Use Learning on the Job: Continual Learning from Deployment Feedback for Frozen-Weights Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.22157; inspect its method and evaluation before treating results as production evidence.
medium
README.md
752
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L752
2026-07-28
Learn
learn
Understand the field and its boundaries.
context
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22157
[2607.22157] Learning on the Job: Continual Learning from Deployment Feedback for Frozen-Weights Agents
AI agents encounter learning opportunities in every episode they run, and discard nearly all of them: the underlying models are frozen at deployment, so an agent that resolves a difficult request today starts from zero when it recurs tomorrow. Yet ordinary operation already produces feedback, in the form of outcome ver...
Valentin Tablan; Scott Taylor; Kristoffer Bernhem
2026-07-24
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.22157
2026-08-07T12:31:05
ale-0174
Research Foundations
research-foundations
Paper
📄
Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills
https://arxiv.org/abs/2607.22529
external
arxiv.org
Names the central dilemma of self-evolving training loops: environment-bound methods get precise feedback but stay narrow, while open-ended self-generation broadens the task space and loses reliable verification, letting misleading rewards pollute the training loop. Positions agent skills as the middle ground -- each s...
Names the central dilemma of self-evolving training loops: environment-bound methods get precise feedback but stay narrow, while open-ended self-generation broadens the task space and loses reliable verification, letting misleading rewards pollute the training loop. Positions agent skills as the middle ground -- each s...
Names the central dilemma of self-evolving training loops: environment-bound methods get precise feedback but stay narrow, while open-ended self-generation broadens the task space and loses reliable verification, letting misleading rewards pollute the training loop. Positions agent skills as the middle ground -- each s...
Verification is promoted from a final check to a loop-control signal. Names the central dilemma of self-evolving training loops: environment-bound methods get precise feedback but stay narrow, while open-ended self-generation broadens the task space and loses reliable verification, letting misleading rewards pollute th...
Use Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.22529; inspect its method and evaluation before treating results as production evidence.
medium
README.md
753
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L753
2026-07-28
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.22529
[2607.22529] Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills
LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fundamental dilemma between task diversity and verification reliability: environment-bound methods obtain precise feedback but confine learning to narrow domains, while ope...
Siyuan Huang; Pengyu Cheng; Haotian Liu; Tao Chen; Yihao Liu; Jingwei Ni; Shijie Zhou; Ziyi Yang; Gangwei Jiang; Mengyu Zhou; Yu Cheng; Xiaoxi Jiang; Guanjun Jiang
2026-07-24
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.22529
2026-08-07T12:31:05
ale-0175
Research Foundations
research-foundations
Paper
📄
Teaching LLMs to Self-Evolve: Cultivating Core Meta-Skills with Reinforcement Learning
https://arxiv.org/abs/2607.21971
external
arxiv.org
Hypothesizes that AlphaEvolve-style test-time self-evolution succeeds because of meta-skills -- notably self-reflection against environment feedback -- that conventional post-training neglects entirely. MetaEvolve cultivates them through a data-synthesis pipeline, evolution-aware RL, and inference-time evolutionary sea...
Hypothesizes that AlphaEvolve-style test-time self-evolution succeeds because of meta-skills -- notably self-reflection against environment feedback -- that conventional post-training neglects entirely. MetaEvolve cultivates them through a data-synthesis pipeline, evolution-aware RL, and inference-time evolutionary sea...
Hypothesizes that AlphaEvolve-style test-time self-evolution succeeds because of meta-skills -- notably self-reflection against environment feedback -- that conventional post-training neglects entirely. MetaEvolve cultivates them through a data-synthesis pipeline, evolution-aware RL, and inference-time evolutionary sea...
Connects Loop Engineering to prior agent-loop and feedback-loop research. Hypothesizes that AlphaEvolve-style test-time self-evolution succeeds because of meta-skills -- notably self-reflection against environment feedback -- that conventional post-training neglects entirely. MetaEvolve cultivates them through a data-s...
Use Teaching LLMs to Self-Evolve: Cultivating Core Meta-Skills with Reinforcement Learning to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.21971; inspect its method and evaluation before treating results as production evidence.
medium
README.md
754
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L754
2026-07-28
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.21971
[2607.21971] Teaching LLMs to Self-Evolve: Cultivating Core Meta-Skills with Reinforcement Learning
Test-time scaling through iterative self-evolution with environment feedback, as demonstrated by AlphaEvolve, shows remarkable performance gains. We hypothesize that the success of such evolution frameworks hinges on meta-skills, such as self-reflection with environment feedback, that enable effective multi-round refin...
Shujin Wu; Cheng Qian; Xiusi Chen; Heng Ji
2026-07-24
2026
arXiv
arXiv
cs.LG
arxiv-api
2607.21971
2026-08-07T12:31:05
ale-0176
Research Foundations
research-foundations
Paper
📄
From Execution to Capability: Scientific Experience Consolidation via Procedural Knowledge Synthesis
https://arxiv.org/abs/2607.24459
external
arxiv.org
Studies why executable feedback on one task rarely becomes durable capability on the next, and names the two obstacles precisely: trajectory-derived artifacts often encode source-specific repairs rather than cross-task mechanisms, and a weaker target model may not be able to operationalize a valid abstract procedure (t...
Studies why executable feedback on one task rarely becomes durable capability on the next, and names the two obstacles precisely: trajectory-derived artifacts often encode source-specific repairs rather than cross-task mechanisms, and a weaker target model may not be able to operationalize a valid abstract procedure (t...
Studies why executable feedback on one task rarely becomes durable capability on the next, and names the two obstacles precisely: trajectory-derived artifacts often encode source-specific repairs rather than cross-task mechanisms, and a weaker target model may not be able to operationalize a valid abstract procedure (t...
Durable execution and replay are treated as first-class loop infrastructure. Studies why executable feedback on one task rarely becomes durable capability on the next, and names the two obstacles precisely: trajectory-derived artifacts often encode source-specific repairs rather than cross-task mechanisms, and a weaker...
Use From Execution to Capability: Scientific Experience Consolidation via Procedural Knowledge Synthesis to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.24459; inspect its method and evaluation before treating results as production evidence.
medium
README.md
755
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L755
2026-07-28
Learn
learn
Understand the field and its boundaries.
verification
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.24459
[2607.24459] From Execution to Capability: Scientific Experience Consolidation via Procedural Knowledge Synthesis
Large language models increasingly solve scientific-computing tasks, but executable feedback from one problem rarely becomes durable capability on subsequent problems. We study scientific-computing experience consolidation: converting verified runtime experience into transferable procedural knowledge and persistent mod...
Liwei Dong; Jiahao Zhao; Nan Xu
2026-07-27
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.24459
2026-08-07T12:31:05
ale-0177
Research Foundations
research-foundations
Paper
📄
Agent-UCT: Upper Confidence Bounds Applied to Trees for Agentic Workflow Optimization with Cost-Awareness
https://arxiv.org/abs/2607.24162
external
arxiv.org
Scaffold optimization as tree search: extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph, biasing selection toward branches that reuse already-materialized configuration prefixes so the search stops re-executing shared pipeline stages under tight evaluation budgets. The RAGSp...
Scaffold optimization as tree search: extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph, biasing selection toward branches that reuse already-materialized configuration prefixes so the search stops re-executing shared pipeline stages under tight evaluation budgets. The RAGSp...
Scaffold optimization as tree search: extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph, biasing selection toward branches that reuse already-materialized configuration prefixes so the search stops re-executing shared pipeline stages under tight evaluation budgets. The RAGSp...
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Scaffold optimization as tree search: extends UCT with a reuse-aware regularization term derived from a bipartite prefix reuse graph, biasing selection toward branches that reuse already-materialized configuration prefixes so the sea...
Use Agent-UCT: Upper Confidence Bounds Applied to Trees for Agentic Workflow Optimization with Cost-Awareness to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.24162; inspect its method and evaluation before treating results as production evidence.
medium
README.md
756
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L756
2026-07-28
Learn
learn
Understand the field and its boundaries.
verification;budget;exit
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.24162
[2607.24162] Agent-UCT: Upper Confidence Bounds Applied to Trees for Agentic Workflow Optimization with Cost-Awareness
Optimizing agentic workflows, such as retrieval-augmented generation (RAG) pipelines, requires navigating a combinatorial space of discrete component choices under tight evaluation budgets. Existing approaches - heuristic search, black-box optimization, and standard tree search methods - do not explicitly exploit the c...
Yang Li; Hai Liu; Dian Shao; Yu Wang; Xiyu Chen; Sergey Volkov; Bozhi Wang; Ziyu Sun; Sihang Liu; Ye Luo; Xiaowei Zhang
2026-07-27
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.24162
2026-08-07T12:31:05
ale-0178
Research Foundations
research-foundations
Paper
📄
The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
https://arxiv.org/abs/2607.24720
external
arxiv.org
Builds a unified controlled multi-turn environment to study where long-horizon planning ability actually comes from, which opaque internet-scale training makes impossible to isolate. Findings usable by loop designers: explicit world-model construction via chain-of-thought state-transition modeling yields stronger long-...
Builds a unified controlled multi-turn environment to study where long-horizon planning ability actually comes from, which opaque internet-scale training makes impossible to isolate. Findings usable by loop designers: explicit world-model construction via chain-of-thought state-transition modeling yields stronger long-...
Builds a unified controlled multi-turn environment to study where long-horizon planning ability actually comes from, which opaque internet-scale training makes impossible to isolate. Findings usable by loop designers: explicit world-model construction via chain-of-thought state-transition modeling yields stronger long-...
The work targets tasks that exceed a single context window or prompt session. Builds a unified controlled multi-turn environment to study where long-horizon planning ability actually comes from, which opaque internet-scale training makes impossible to isolate. Findings usable by loop designers: explicit world-model con...
Use The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.24720; inspect its method and evaluation before treating results as production evidence.
medium
README.md
757
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L757
2026-07-28
Learn
learn
Understand the field and its boundaries.
state
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.24720
[2607.24720] The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation
Multi-turn long-horizon planning is critical for foundation model agents, yet how to fundamentally improve it remains unclear. Existing models are trained on uncontrollable and opaque Internet data, making it difficult to identify how planning ability is acquired, shaped, and integrated. To address this challenge, we i...
Tianyi Men; Zhuoran Jin; Kang Liu; Jun Zhao
2026-07-27
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.24720
2026-08-07T12:31:05
ale-0179
Research Foundations
research-foundations
Paper
📄
Context Assembly as the Controlled Variable: A Control-Theoretic View of Harness Policies for Frozen LLM Agents
https://arxiv.org/abs/2607.25408
external
arxiv.org
Formalizes the loop as a frozen inner model wrapped by an outer context policy, making prompt templates, demonstrations, and retrieved context the controlled variable, with stability guarantees and uncertainty calibration for online updates. A rare attempt to give harness engineering actual control-theoretic foundation...
Formalizes the loop as a frozen inner model wrapped by an outer context policy, making prompt templates, demonstrations, and retrieved context the controlled variable, with stability guarantees and uncertainty calibration for online updates. A rare attempt to give harness engineering actual control-theoretic foundation...
Formalizes the loop as a frozen inner model wrapped by an outer context policy, making prompt templates, demonstrations, and retrieved context the controlled variable, with stability guarantees and uncertainty calibration for online updates. A rare attempt to give harness engineering actual control-theoretic foundation...
Context is managed as durable loop state rather than a single prompt payload. Formalizes the loop as a frozen inner model wrapped by an outer context policy, making prompt templates, demonstrations, and retrieved context the controlled variable, with stability guarantees and uncertainty calibration for online updates. ...
Use Context Assembly as the Controlled Variable: A Control-Theoretic View of Harness Policies for Frozen LLM Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.25408; inspect its method and evaluation before treating results as production evidence.
medium
README.md
758
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L758
2026-07-30
Learn
learn
Understand the field and its boundaries.
context
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25408
[2607.25408] Context Assembly as the Controlled Variable: A Control-Theoretic View of Harness Policies for Frozen LLM Agents
A growing body of 2026 work applies control theory to LLM agents: Lyapunov-certified stability for tool-mediated controllers (Prinos et al., "Stable Agentic Control", 2026), sample-complexity bounds for sparse policies over massive discrete tool universes (Majumdar, "Sparse Agentic Control", 2026), and regulatory-contr...
Debjyoti Paul
2026-07-28
2026
arXiv
arXiv
6 pages, 2 figures, 1 table. Code and companion paper's data: https://github.com/dpaul0501/context-optimization-rl
cs.AI
arxiv-api
2607.25408
2026-08-07T12:31:05
ale-0180
Research Foundations
research-foundations
Paper
📄
Towards an Agent Operating System - Lessons from Classical and Cloud OS
https://arxiv.org/abs/2607.25076
external
arxiv.org
Argues agentic systems are pre-standardization and proposes deriving abstractions by extending classical OS and cloud primitives to stochastic, natural-language-mediated execution, the way POSIX and Kubernetes consolidated their eras. A useful frame for where loop infrastructure is heading.
Argues agentic systems are pre-standardization and proposes deriving abstractions by extending classical OS and cloud primitives to stochastic, natural-language-mediated execution, the way POSIX and Kubernetes consolidated their eras. A useful frame for where loop infrastructure is heading.
Argues agentic systems are pre-standardization and proposes deriving abstractions by extending classical OS and cloud primitives to stochastic, natural-language-mediated execution, the way POSIX and Kubernetes consolidated their eras. A useful frame for where loop infrastructure is heading.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Argues agentic systems are pre-standardization and proposes deriving abstractions by extending classical OS and cloud primitives to stochastic, natural-language-mediated execution, the way POSIX and Kubernetes consolidated their eras. A useful fr...
Use Towards an Agent Operating System - Lessons from Classical and Cloud OS to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.25076; inspect its method and evaluation before treating results as production evidence.
medium
README.md
759
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L759
2026-07-30
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.25076
[2607.25076] Towards an Agent Operating System - Lessons from Classical and Cloud OS
Every major wave of platform software follows the same arc: an initial period of experimentation with competing frameworks and ad-hoc implementations, followed by the articulation of a small set of stable abstractions with well-defined semantics, and finally consolidation around those abstractions into a platform that ...
Gosia Steinder; Hubertus Franke
2026-07-27
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.25076
2026-08-07T12:31:05
ale-0181
Research Foundations
research-foundations
Blog
📝
Discovering Cryptographic Weaknesses with Claude
https://www.anthropic.com/research/discovering-cryptographic-weaknesses
external
www.anthropic.com
Anthropic research report (2026-07-28) that doubles as one of the most detailed first-party accounts of a multi-day autonomous agent loop.
Anthropic research report (2026-07-28) that doubles as one of the most detailed first-party accounts of a multi-day autonomous agent loop.
Anthropic research report (2026-07-28) that doubles as one of the most detailed first-party accounts of a multi-day autonomous agent loop.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Anthropic research report (2026-07-28) that doubles as one of the most detailed first-party accounts of a multi-day autonomous agent loop.
Use Discovering Cryptographic Weaknesses with Claude to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Contextual source from www.anthropic.com; useful for practice signals or boundary conditions, not independent validation.
contextual
README.md
760
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L760
2026-07-30
Learn
learn
Understand the field and its boundaries.
whole-loop
builder
cross-layer
enabling
practitioner-analysis
B
ok
https://www.anthropic.com/research/discovering-cryptographic-weaknesses
Discovering cryptographic weaknesses with Claude \ Anthropic
Anthropic researchers find weaknesses in cryptographic algorithms with Claude Mythos Preview
Anthropic
domain-fallback
2026-08-07T12:31:05
ale-0182
Research Foundations
research-foundations
Paper
📄
Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering
https://arxiv.org/abs/2607.28568
external
arxiv.org
Builds OpenMLE around four composable operators, Draft, Improve, Debug, Crossover, that compose into long-horizon search strategies, and trains a 35B execution-grounded model on them.
Builds OpenMLE around four composable operators, Draft, Improve, Debug, Crossover, that compose into long-horizon search strategies, and trains a 35B execution-grounded model on them.
Builds OpenMLE around four composable operators, Draft, Improve, Debug, Crossover, that compose into long-horizon search strategies, and trains a 35B execution-grounded model on them.
The work targets tasks that exceed a single context window or prompt session. Builds OpenMLE around four composable operators, Draft, Improve, Debug, Crossover, that compose into long-horizon search strategies, and trains a 35B execution-grounded model on them.
Use Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.28568; inspect its method and evaluation before treating results as production evidence.
medium
README.md
761
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L761
2026-08-05
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.28568
[2607.28568] Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering
Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environment...
Junlin Yang; Che Jiang; Yu Fu; Tianwei Luo; Can Ren; Weizhi Wang; Kaikai Zhao; Hongyi Liu; Yuxin Zuo; Yuru Wang; Yuchen Fan; Kai Tian; Zhenzhao Yuan; Xiaojian Lin; Li Sheng; Rushi Qiang; Guoli Jia; Xingtai Lv; Ermo Hua; Dianqiao Lei; Youbang Sun; Ning Ding; Bowen Zhou; Kaiyan Zhang
2026-07-30
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.28568
2026-08-07T12:31:05
ale-0183
Research Foundations
research-foundations
Paper
📄
Rehearse: Stepping Back from the Confidence Cliff in Self-Improving Autoresearch
https://arxiv.org/abs/2607.27687
external
arxiv.org
Names and measures the 'confidence cliff': as a self-improving autoresearch system accumulates successful modifications, its ability to predict which next idea will work collapses from ~83% to ~57%, so the loop degrades exactly when it looks healthiest.
Names and measures the 'confidence cliff': as a self-improving autoresearch system accumulates successful modifications, its ability to predict which next idea will work collapses from ~83% to ~57%, so the loop degrades exactly when it looks healthiest.
Names and measures the 'confidence cliff': as a self-improving autoresearch system accumulates successful modifications, its ability to predict which next idea will work collapses from ~83% to ~57%, so the loop degrades exactly when it looks healthiest.
Connects Loop Engineering to prior agent-loop and feedback-loop research. Names and measures the 'confidence cliff': as a self-improving autoresearch system accumulates successful modifications, its ability to predict which next idea will work collapses from ~83% to ~57%, so the loop degrades exactly when it looks heal...
Use Rehearse: Stepping Back from the Confidence Cliff in Self-Improving Autoresearch to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.27687; inspect its method and evaluation before treating results as production evidence.
medium
README.md
762
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L762
2026-08-05
Learn
learn
Understand the field and its boundaries.
whole-loop
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.27687
[2607.27687] Rehearse: Stepping Back from the Confidence Cliff in Self-Improving Autoresearch
Autoresearch improves machine-learning code by proposing changes, running full training jobs, and keeping changes that improve the metric. The efficiency of this loop depends not only on generating ideas, but also on the agent's ability to decide, before spending a training run, whether a proposed modification is likel...
Jiazhen Ji; Shouhong Ding
2026-07-30
2026
arXiv
arXiv
cs.AI
arxiv-api
2607.27687
2026-08-07T12:31:05
ale-0184
Research Foundations
research-foundations
Paper
📄
Harness-G: A Graph-Structured Harness for Search Agents
https://arxiv.org/abs/2607.27652
external
arxiv.org
Diagnoses 'retrieval-equivalence collapse' in RL-trained search agents: during Search-R1 training, rollouts for the same question keep emitting distinct query strings while their accumulated evidence sets increasingly overlap, so trajectories converge to utility equivalence and within-group returns carry almost no retr...
Diagnoses 'retrieval-equivalence collapse' in RL-trained search agents: during Search-R1 training, rollouts for the same question keep emitting distinct query strings while their accumulated evidence sets increasingly overlap, so trajectories converge to utility equivalence and within-group returns carry almost no retr...
Diagnoses 'retrieval-equivalence collapse' in RL-trained search agents: during Search-R1 training, rollouts for the same question keep emitting distinct query strings while their accumulated evidence sets increasingly overlap, so trajectories converge to utility equivalence and within-group returns carry almost no retr...
Control flow is represented as an inspectable graph rather than an opaque prompt loop. Diagnoses 'retrieval-equivalence collapse' in RL-trained search agents: during Search-R1 training, rollouts for the same question keep emitting distinct query strings while their accumulated evidence sets increasingly overlap, so tra...
Use Harness-G: A Graph-Structured Harness for Search Agents to understand the evidence, vocabulary, and lineage behind recurring agent systems.
Research source arXiv:2607.27652; inspect its method and evaluation before treating results as production evidence.
medium
README.md
763
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L763
2026-08-07
Learn
learn
Understand the field and its boundaries.
context
researcher;evaluator
cross-layer
enabling
research-preprint
A
ok
https://arxiv.org/abs/2607.27652
[2607.27652] Harness-G: A Graph-Structured Harness for Search Agents
Reinforcement learning (RL) search agents commonly model retrieval as free-form natural-language query generation and optimize multi-turn interactions using final-answer rewards. Current studies mainly improve training with denser or more structured credit signals, but rarely examine whether retrieval is properly formu...
Yanning Hou; Haoyuan Chen; Sihang Zhou; Xiaoshu Chen; Xirui Liu; Duanyang Yuan; Lingyuan Meng; Siwei Wang; Quan Liu; Jian Huang
2026-07-30
2026
arXiv
arXiv
Code:https://github.com/7HHHHH/Harness-G
cs.CL
arxiv-api
2607.27652
2026-08-07T12:31:05
ale-0185
Model-Level Recurrence
model-level-recurrence
Paper
📄
Universal Transformers
https://openreview.net/forum?id=HyzdRiR9Y7
external
openreview.net
Introduces recurrent depth for Transformers by repeatedly applying shared self-attention and transition blocks, with optional per-position adaptive halting; establishes the architectural foundation for later looped models.
Introduces recurrent depth for Transformers by repeatedly applying shared self-attention and transition blocks, with optional per-position adaptive halting; establishes the architectural foundation for later looped models.
Introduces recurrent depth for Transformers by repeatedly applying shared self-attention and transition blocks, with optional per-position adaptive halting; establishes the architectural foundation for later looped models.
Reuses learned computation inside one model inference rather than repeating a full agent run. Introduces recurrent depth for Transformers by repeatedly applying shared self-attention and transition blocks, with optional per-position adaptive halting; establishes the architectural foundation for later looped models.
Use Universal Transformers to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
783
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L783
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/challenge?redirect=%2Fforum%3Fid%3DHyzdRiR9Y7
Verifying your browser | OpenReview
Mostafa Dehghani; Stephan Gouws; Oriol Vinyals; Jakob Uszkoreit; Łukasz Kaiser
2019
2019
International Conference on Learning Representations (ICLR)
OpenReview
Published at ICLR 2019; venue and authors verified from the official OpenReview record.
OpenReview
2026-08-07T12:31:05
ale-0186
Model-Level Recurrence
model-level-recurrence
Paper
📄
Looped Transformers as Programmable Computers
https://proceedings.mlr.press/v202/giannou23a.html
external
proceedings.mlr.press
Constructs a constant-depth looped Transformer that advances an in-state program counter and executes reusable instructions, showing how iterative algorithms and in-context gradient descent can be represented through repeated shared computation.
Constructs a constant-depth looped Transformer that advances an in-state program counter and executes reusable instructions, showing how iterative algorithms and in-context gradient descent can be represented through repeated shared computation.
Constructs a constant-depth looped Transformer that advances an in-state program counter and executes reusable instructions, showing how iterative algorithms and in-context gradient descent can be represented through repeated shared computation.
Reuses learned computation inside one model inference rather than repeating a full agent run. Constructs a constant-depth looped Transformer that advances an in-state program counter and executes reusable instructions, showing how iterative algorithms and in-context gradient descent can be represented through repeated ...
Use Looped Transformers as Programmable Computers to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
784
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L784
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;context;state
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://proceedings.mlr.press/v202/giannou23a.html
Looped Transformers as Programmable Computers
Looped Transformers as Programmable ComputersAngeliki Giannou, Shashank Rajput, Jy-Yong Sohn, Kangwook Lee, Jason D. Lee, Dimitris P...
Angeliki Giannou; Shashank Rajput; Jy-Yong Sohn; Kangwook Lee; Jason D. Lee; Dimitris Papailiopoulos
2023-07-03
2023
International Conference on Machine Learning
PMLR
html-meta
2026-08-07T12:31:05
ale-0187
Model-Level Recurrence
model-level-recurrence
Paper
📄
Looped Transformers are Better at Learning Learning Algorithms
https://openreview.net/forum?id=HHbRxoDTxE
external
openreview.net
Trains input-injected looped Transformers for in-context data fitting and shows that iterative shared computation can match standard Transformers on tested function classes with substantially fewer parameters.
Trains input-injected looped Transformers for in-context data fitting and shows that iterative shared computation can match standard Transformers on tested function classes with substantially fewer parameters.
Trains input-injected looped Transformers for in-context data fitting and shows that iterative shared computation can match standard Transformers on tested function classes with substantially fewer parameters.
Reuses learned computation inside one model inference rather than repeating a full agent run. Trains input-injected looped Transformers for in-context data fitting and shows that iterative shared computation can match standard Transformers on tested function classes with substantially fewer parameters.
Use Looped Transformers are Better at Learning Learning Algorithms to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
785
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L785
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;context;verification
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/challenge?redirect=%2Fforum%3Fid%3DHHbRxoDTxE
Verifying your browser | OpenReview
Liu Yang; Kangwook Lee; Robert D. Nowak; Dimitris Papailiopoulos
2024
2024
International Conference on Learning Representations (ICLR)
OpenReview
Published at ICLR 2024; venue and authors verified from the official OpenReview record.
OpenReview
2026-08-07T12:31:05
ale-0188
Model-Level Recurrence
model-level-recurrence
Paper
📄
On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encoding
https://proceedings.mlr.press/v267/xu25x.html
external
proceedings.mlr.press
Derives approximation rates for looped Transformers, identifies a loop-specific expressivity limit, and uses timestep-conditioned scaling to improve function approximation as recurrence increases.
Derives approximation rates for looped Transformers, identifies a loop-specific expressivity limit, and uses timestep-conditioned scaling to improve function approximation as recurrence increases.
Derives approximation rates for looped Transformers, identifies a loop-specific expressivity limit, and uses timestep-conditioned scaling to improve function approximation as recurrence increases.
Reuses learned computation inside one model inference rather than repeating a full agent run. Derives approximation rates for looped Transformers, identifies a loop-specific expressivity limit, and uses timestep-conditioned scaling to improve function approximation as recurrence increases.
Use On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encoding to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
786
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L786
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://proceedings.mlr.press/v267/xu25x.html
On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encoding
On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep EncodingKevin Xu, Issei SatoLooped Transformers provide ad...
Kevin Xu; Issei Sato
2025-10-06
2025
International Conference on Machine Learning
PMLR
html-meta
2026-08-07T12:31:05
ale-0189
Model-Level Recurrence
model-level-recurrence
Paper
📄
Reasoning with Latent Thoughts: On the Power of Looped Transformers
https://iclr.cc/virtual/2025/poster/28971
external
iclr.cc
Connects effective recurrent depth to reasoning, proves that looped models can simulate multi-step chain-of-thought in latent space under the paper's construction, and studies the trade-off between reasoning and memorization.
Connects effective recurrent depth to reasoning, proves that looped models can simulate multi-step chain-of-thought in latent space under the paper's construction, and studies the trade-off between reasoning and memorization.
Connects effective recurrent depth to reasoning, proves that looped models can simulate multi-step chain-of-thought in latent space under the paper's construction, and studies the trade-off between reasoning and memorization.
Reuses learned computation inside one model inference rather than repeating a full agent run. Connects effective recurrent depth to reasoning, proves that looped models can simulate multi-step chain-of-thought in latent space under the paper's construction, and studies the trade-off between reasoning and memorization.
Use Reasoning with Latent Thoughts: On the Power of Looped Transformers to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
787
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L787
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://iclr.cc/virtual/2025/poster/28971
ICLR Poster Reasoning with Latent Thoughts: On the Power of Looped Transformers ICLR 2025
Nikunj Saunshi; Nishanth Dikkala; Zhiyuan Li; Sanjiv Kumar; Sashank J. Reddi
2025
2025
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published at ICLR 2025; metadata verified from the official conference poster page.
ICLR proceedings
2026-08-07T12:31:05
ale-0190
Model-Level Recurrence
model-level-recurrence
Paper
📄
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
https://arxiv.org/abs/2502.05171
external
arxiv.org
Presents Huginn, a 3.5B recurrent-depth language model trained on 800B tokens whose shared core can be unrolled further at inference, with gains concentrated on reasoning tasks and support for adaptive compute and KV-cache sharing.
Presents Huginn, a 3.5B recurrent-depth language model trained on 800B tokens whose shared core can be unrolled further at inference, with gains concentrated on reasoning tasks and support for adaptive compute and KV-cache sharing.
Presents Huginn, a 3.5B recurrent-depth language model trained on 800B tokens whose shared core can be unrolled further at inference, with gains concentrated on reasoning tasks and support for adaptive compute and KV-cache sharing.
Reuses learned computation inside one model inference rather than repeating a full agent run. Presents Huginn, a 3.5B recurrent-depth language model trained on 800B tokens whose shared core can be unrolled further at inference, with gains concentrated on reasoning tasks and support for adaptive compute and KV-cache sha...
Use Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2502.05171; inspect its method and evaluation before treating results as production evidence.
medium
README.md
793
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L793
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;verification;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/forum?id=D6o6Bwtq7h
[2502.05171] Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach
We study a novel language model architecture that is capable of scaling test-time computation by implicitly reasoning in latent space. Our model works by iterating a recurrent block, thereby unrolling to arbitrary depth at test-time. This stands in contrast to mainstream reasoning models that scale up compute by produc...
Jonas Geiping; Sean McLeish; Neel Jain; John Kirchenbauer; Siddharth Singh; Brian R. Bartoldson; Bhavya Kailkhura; Abhinav Bhatele; Tom Goldstein
2025
2025
Advances in Neural Information Processing Systems 38 (NeurIPS 2025)
Neural Information Processing Systems Foundation
Published in Advances in Neural Information Processing Systems 38 (NeurIPS 2025); the linked arXiv record remains available for open access.
cs.LG
OpenReview conference record
2502.05171
2026-08-07T12:31:05
ale-0191
Model-Level Recurrence
model-level-recurrence
Paper
📄
Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
https://arxiv.org/abs/2507.10524
external
arxiv.org
Combines shared recursive layers with token-level routers so difficult tokens receive more depth while attention and KV caching are restricted to active tokens.
Combines shared recursive layers with token-level routers so difficult tokens receive more depth while attention and KV caching are restricted to active tokens.
Combines shared recursive layers with token-level routers so difficult tokens receive more depth while attention and KV caching are restricted to active tokens.
Reuses learned computation inside one model inference rather than repeating a full agent run. Combines shared recursive layers with token-level routers so difficult tokens receive more depth while attention and KV caching are restricted to active tokens.
Use Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2507.10524; inspect its method and evaluation before treating results as production evidence.
medium
README.md
794
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L794
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/forum?id=QuqsEIVWIG
[2507.10524] Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
Scaling language models unlocks impressive capabilities, but the accompanying computational and memory demands make both training and deployment expensive. Existing efficiency efforts typically target either parameter sharing or adaptive computation, leaving open the question of how to attain both simultaneously. We in...
Sangmin Bae; Yujin Kim; Reza Bayat; Sungnyun Kim; Jiyoun Ha; Tal Schuster; Adam Fisch; Hrayr Harutyunyan; Ziwei Ji; Aaron Courville; Se-Young Yun
2025
2025
Advances in Neural Information Processing Systems 38 (NeurIPS 2025)
Neural Information Processing Systems Foundation
Published in Advances in Neural Information Processing Systems 38 (NeurIPS 2025); the linked arXiv record remains available for open access.
cs.CL
OpenReview conference record
2507.10524
2026-08-07T12:31:05
ale-0192
Model-Level Recurrence
model-level-recurrence
Paper
📄
Scaling Latent Reasoning via Looped Language Models
https://arxiv.org/abs/2510.25741
external
arxiv.org
Introduces the Ouro family of pretrained LoopLMs, combining latent iteration, learned depth allocation, and large-scale pretraining to study recurrent depth as a scaling axis distinct from parameter count and generated reasoning tokens.
Introduces the Ouro family of pretrained LoopLMs, combining latent iteration, learned depth allocation, and large-scale pretraining to study recurrent depth as a scaling axis distinct from parameter count and generated reasoning tokens.
Introduces the Ouro family of pretrained LoopLMs, combining latent iteration, learned depth allocation, and large-scale pretraining to study recurrent depth as a scaling axis distinct from parameter count and generated reasoning tokens.
Reuses learned computation inside one model inference rather than repeating a full agent run. Introduces the Ouro family of pretrained LoopLMs, combining latent iteration, learned depth allocation, and large-scale pretraining to study recurrent depth as a scaling axis distinct from parameter count and generated reasoni...
Use Scaling Latent Reasoning via Looped Language Models to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2510.25741; inspect its method and evaluation before treating results as production evidence.
medium
README.md
795
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L795
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-preprint
A
ok
https://arxiv.org/abs/2510.25741
[2510.25741] Scaling Latent Reasoning via Looped Language Models
Modern LLMs are trained to "think" primarily via explicit text generation, such as chain-of-thought (CoT), which defers reasoning to post-training and under-leverages pre-training data. We present and open-source Ouro, named after the recursive Ouroboros, a family of pre-trained Looped Language Models (LoopLM) that ins...
Rui-Jie Zhu; Zixuan Wang; Kai Hua; Tianyu Zhang; Ziniu Li; Haoran Que; Boyi Wei; Zixin Wen; Fan Yin; He Xing; Lu Li; Jiajun Shi; Kaijing Ma; Shanda Li; Taylor Kergan; Andrew Smith; Xingwei Qu; Mude Hui; Bohong Wu; Qiyang Min; Hongzhi Huang; Xun Zhou; Wei Ye; Jiaheng Liu; Jian Yang; Yunfeng Shi; Chenghua Lin; Enduo Zhao...
2025-10-29
2025
arXiv
arXiv
cs.CL
arxiv-api
2510.25741
2026-08-07T12:31:05
ale-0193
Model-Level Recurrence
model-level-recurrence
Paper
📄
LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation
https://iclr.cc/virtual/2026/poster/10009450
external
iclr.cc
Trains variable-length latent trajectories with time and step-size conditioning plus shortcut consistency, allowing one model to trade compute for quality across inference budgets without retraining.
Trains variable-length latent trajectories with time and step-size conditioning plus shortcut consistency, allowing one model to trade compute for quality across inference budgets without retraining.
Trains variable-length latent trajectories with time and step-size conditioning plus shortcut consistency, allowing one model to trade compute for quality across inference budgets without retraining.
Reuses learned computation inside one model inference rather than repeating a full agent run. Trains variable-length latent trajectories with time and step-size conditioning plus shortcut consistency, allowing one model to trade compute for quality across inference budgets without retraining.
Use LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
796
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L796
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://iclr.cc/virtual/2026/poster/10009450
ICLR Poster LoopFormer: Elastic-Depth Looped Transformers for Latent Reasoning via Shortcut Modulation ICLR 2026
Ahmadreza Jeddi; Marco Ciccone; Babak Taati
2026
2026
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published at ICLR 2026; metadata verified from the official conference poster page.
ICLR proceedings
2026-08-07T12:31:05
ale-0194
Model-Level Recurrence
model-level-recurrence
Paper
📄
MoDr: Mixture-of-Depth-Recurrent Transformers for Test-Time Reasoning
https://iclr.cc/virtual/2026/poster/10011117
external
iclr.cc
Replaces a single recurrent reasoning path with dynamically routed LoRA branches, adding solution-space exploration and load-balanced routing to the Huginn-style depth-recurrent backbone.
Replaces a single recurrent reasoning path with dynamically routed LoRA branches, adding solution-space exploration and load-balanced routing to the Huginn-style depth-recurrent backbone.
Replaces a single recurrent reasoning path with dynamically routed LoRA branches, adding solution-space exploration and load-balanced routing to the Huginn-style depth-recurrent backbone.
Reuses learned computation inside one model inference rather than repeating a full agent run. Replaces a single recurrent reasoning path with dynamically routed LoRA branches, adding solution-space exploration and load-balanced routing to the Huginn-style depth-recurrent backbone.
Use MoDr: Mixture-of-Depth-Recurrent Transformers for Test-Time Reasoning to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
797
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L797
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;verification
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://iclr.cc/virtual/2026/poster/10011117
ICLR Poster MoDr: Mixture-of-Depth-Recurrent Transformers for Test-Time Reasoning ICLR 2026
Xiaojing Zhang; Haifeng Wu; Gang He; Jiyang Shen; Bochen Lyu; Zhanxing Zhu
2026
2026
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published at ICLR 2026; metadata verified from the official conference poster page.
ICLR proceedings
2026-08-07T12:31:05
ale-0195
Model-Level Recurrence
model-level-recurrence
Paper
📄
ChainGPT: Dual-Reasoning Model with Recurrent Depth and Multi-Rank State Updates
https://iclr.cc/virtual/2026/poster/10007767
external
iclr.cc
Combines within-layer multi-substep state updates, state-guided sparse attention, across-layer recurrence, and adaptive stopping to increase latent reasoning depth without extending visible chain-of-thought.
Combines within-layer multi-substep state updates, state-guided sparse attention, across-layer recurrence, and adaptive stopping to increase latent reasoning depth without extending visible chain-of-thought.
Combines within-layer multi-substep state updates, state-guided sparse attention, across-layer recurrence, and adaptive stopping to increase latent reasoning depth without extending visible chain-of-thought.
Reuses learned computation inside one model inference rather than repeating a full agent run. Combines within-layer multi-substep state updates, state-guided sparse attention, across-layer recurrence, and adaptive stopping to increase latent reasoning depth without extending visible chain-of-thought.
Use ChainGPT: Dual-Reasoning Model with Recurrent Depth and Multi-Rank State Updates to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
798
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L798
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;state;exit
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://iclr.cc/virtual/2026/poster/10007767
ICLR Poster ChainGPT: Dual-Reasoning Model with Recurrent Depth and Multi-Rank State Updates ICLR 2026
Yunao Zheng; Xiaojie Wang; Lei Ren; Chen Wei
2026
2026
International Conference on Learning Representations (ICLR)
International Conference on Learning Representations
Published at ICLR 2026; metadata verified from the official conference poster page.
ICLR proceedings
2026-08-07T12:31:05
ale-0196
Model-Level Recurrence
model-level-recurrence
Paper
📄
Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models
https://openreview.net/forum?id=eQaJSRZiGn
external
openreview.net
Learns when a token needs extra latent refinement, using a neural decider, depth-aware LoRA, and cross-iteration attention to avoid always paying for or being degraded by additional loops.
Learns when a token needs extra latent refinement, using a neural decider, depth-aware LoRA, and cross-iteration attention to avoid always paying for or being degraded by additional loops.
Learns when a token needs extra latent refinement, using a neural decider, depth-aware LoRA, and cross-iteration attention to avoid always paying for or being degraded by additional loops.
Reuses learned computation inside one model inference rather than repeating a full agent run. Learns when a token needs extra latent refinement, using a neural decider, depth-aware LoRA, and cross-iteration attention to avoid always paying for or being degraded by additional loops.
Use Think-at-Hard: Selective Latent Iterations to Improve Reasoning Language Models to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source; inspect its method and evaluation before treating results as production evidence.
medium
README.md
799
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L799
2026-07-18
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/challenge?redirect=%2Fforum%3Fid%3DeQaJSRZiGn
Verifying your browser | OpenReview
Tianyu Fu; Yichen You; Zekai Chen; Guohao Dai; Huazhong Yang; Yu Wang
2026
2026
International Conference on Machine Learning (ICML)
OpenReview
Published at ICML 2026; venue and authors verified from the official OpenReview record.
OpenReview
2026-08-07T12:31:05
ale-0197
Model-Level Recurrence
model-level-recurrence
Paper
📄
Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers
https://arxiv.org/abs/2606.18206
external
arxiv.org
Stabilizes very deep recurrence with pre-normalization and residual scaling, then uses latent-state convergence as the halting signal so the model allocates more iterations to harder Sudoku, maze, state-tracking, and ARC-AGI instances without a separate stopping head.
Stabilizes very deep recurrence with pre-normalization and residual scaling, then uses latent-state convergence as the halting signal so the model allocates more iterations to harder Sudoku, maze, state-tracking, and ARC-AGI instances without a separate stopping head.
Stabilizes very deep recurrence with pre-normalization and residual scaling, then uses latent-state convergence as the halting signal so the model allocates more iterations to harder Sudoku, maze, state-tracking, and ARC-AGI instances without a separate stopping head.
Reuses learned computation inside one model inference rather than repeating a full agent run. Stabilizes very deep recurrence with pre-normalization and residual scaling, then uses latent-state convergence as the halting signal so the model allocates more iterations to harder Sudoku, maze, state-tracking, and ARC-AGI i...
Use Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2606.18206; inspect its method and evaluation before treating results as production evidence.
medium
README.md
800
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L800
2026-07-20
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;state;exit
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-paper
A
ok
https://openreview.net/pdf/51350b6e425ed0500ac9eb9cec78ba15d9f5d1ba.pdf
[2606.18206] Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers
Looped architectures provide an inductive bias toward learning step-by-step procedures for tasks that require compositional reasoning. The number of effective layers reached by looping determines the quality of the solution these models find. Like deep architectures, looped architectures are prone to a signal propagati...
Sajad Movahedi; Vera Milovanović; Shlomo Libo Feigin; Alexander Theus; Thomas Hofmann; Valentina Boeva; T. Konstantin Rusch; Antonio Orvieto
2026
2026
Proceedings of the 43rd International Conference on Machine Learning (ICML), PMLR 306
PMLR
Published in Proceedings of the 43rd International Conference on Machine Learning (ICML), PMLR 306; the linked arXiv record remains available for open access.
cs.AI
PMLR camera-ready record
2606.18206
2026-08-07T12:31:05
ale-0198
Model-Level Recurrence
model-level-recurrence
Paper
📄
Loop the Loopies!
https://arxiv.org/abs/2607.16051
external
arxiv.org
Introduces the Loopie family of sparse looped Transformers (20B parameters with 2B active and 6B with 0.6B active); the authors' matched-compute experiments over 3.5T pretraining tokens overtake a reproduced vanilla 30B-A3B baseline after roughly 600B tokens, while post-training plus test-time scaling reaches 35/42 on ...
Introduces the Loopie family of sparse looped Transformers (20B parameters with 2B active and 6B with 0.6B active); the authors' matched-compute experiments over 3.5T pretraining tokens overtake a reproduced vanilla 30B-A3B baseline after roughly 600B tokens, while post-training plus test-time scaling reaches 35/42 on ...
Introduces the Loopie family of sparse looped Transformers (20B parameters with 2B active and 6B with 0.6B active); the authors' matched-compute experiments over 3.5T pretraining tokens overtake a reproduced vanilla 30B-A3B baseline after roughly 600B tokens, while post-training plus test-time scaling reaches 35/42 on ...
Reuses learned computation inside one model inference rather than repeating a full agent run. Introduces the Loopie family of sparse looped Transformers (20B parameters with 2B active and 6B with 0.6B active); the authors' matched-compute experiments over 3.5T pretraining tokens overtake a reproduced vanilla 30B-A3B ba...
Use Loop the Loopies! to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2607.16051; inspect its method and evaluation before treating results as production evidence.
medium
README.md
806
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L806
2026-07-20
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;verification;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-preprint
A
ok
https://arxiv.org/abs/2607.16051
[2607.16051] Loop the Loopies!
We present the Loopie series, consisting of two Mixture-of-Experts (MoE) models: a 20B-parameter model with 2B active parameters and a 6B-parameter model with 0.6B active parameters. Looped Transformers have long faced a challenge: given an N times increase in pre-training compute, increasing the parameter count by a f...
Zitian Gao; Yilong Chen; Yihao Xiao; Xinyu Yang; Ran Tao; Joey Zhou; Bryan Dai
2026-07-17
2026
arXiv
arXiv
cs.CL
arxiv-api
2607.16051
2026-08-07T12:31:05
ale-0199
Model-Level Recurrence
model-level-recurrence
Paper
📄
LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling
https://arxiv.org/abs/2606.04438
external
arxiv.org
Combines sparse expert routing with weight-shared recurrence through iteration-conditioned adaptive normalization and capacity balancing; under matched parameters, per-token FLOPs, and active-sublayer ratios, the reported 3B model beats its vanilla MoE control on 8 of 9 benchmarks by more than one point on average, wit...
Combines sparse expert routing with weight-shared recurrence through iteration-conditioned adaptive normalization and capacity balancing; under matched parameters, per-token FLOPs, and active-sublayer ratios, the reported 3B model beats its vanilla MoE control on 8 of 9 benchmarks by more than one point on average, wit...
Combines sparse expert routing with weight-shared recurrence through iteration-conditioned adaptive normalization and capacity balancing; under matched parameters, per-token FLOPs, and active-sublayer ratios, the reported 3B model beats its vanilla MoE control on 8 of 9 benchmarks by more than one point on average, wit...
Reuses learned computation inside one model inference rather than repeating a full agent run. Combines sparse expert routing with weight-shared recurrence through iteration-conditioned adaptive normalization and capacity balancing; under matched parameters, per-token FLOPs, and active-sublayer ratios, the reported 3B m...
Use LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2606.04438; inspect its method and evaluation before treating results as production evidence.
medium
README.md
807
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L807
2026-07-20
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;verification;budget
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-preprint
A
ok
https://arxiv.org/abs/2606.04438
[2606.04438] LoopMoE: Unifying Iterative Computation with Mixture-of-Experts for Language Modeling
Mixture-of-Experts (MoE) and looped architectures scale models along two orthogonal axes, namely parameter capacity and effective depth. However, mainstream looped architectures rely on dense backbones that couple parameter count with per-token FLOPs, which makes it impossible to isolate the effect of iterative computa...
Wenkai Chen; Tianshu Li; Wenyong Huang; Yichun Yin; Lifeng Shang; Chengwei Qin
2026-06-03
2026
arXiv
arXiv
cs.LG
arxiv-api
2606.04438
2026-08-07T12:31:05
ale-0200
Model-Level Recurrence
model-level-recurrence
Paper
📄
Sparse Layers are Critical to Scaling Looped Language Models
https://arxiv.org/abs/2605.09165
external
arxiv.org
Finds that dense looped models lose their scaling advantage while looped MoE models recover expressivity because routing diverges across repeated passes; also shows that loop boundaries form stronger early-exit points than arbitrary layers in standard Transformers.
Finds that dense looped models lose their scaling advantage while looped MoE models recover expressivity because routing diverges across repeated passes; also shows that loop boundaries form stronger early-exit points than arbitrary layers in standard Transformers.
Finds that dense looped models lose their scaling advantage while looped MoE models recover expressivity because routing diverges across repeated passes; also shows that loop boundaries form stronger early-exit points than arbitrary layers in standard Transformers.
Reuses learned computation inside one model inference rather than repeating a full agent run. Finds that dense looped models lose their scaling advantage while looped MoE models recover expressivity because routing diverges across repeated passes; also shows that loop boundaries form stronger early-exit points than arb...
Use Sparse Layers are Critical to Scaling Looped Language Models to assess inner latent computation as a model capability inside a separately governed agent loop.
Research source arXiv:2605.09165; inspect its method and evaluation before treating results as production evidence.
medium
README.md
808
https://github.com/ChaoYue0307/awesome-loop-engineering/blob/main/README.md#L808
2026-07-20
Learn
learn
Understand how recurrent model computation can power, but not replace, a governed agent loop.
act;exit
researcher;evaluator;model-builder;agent-builder
model
adjacent
research-preprint
A
ok
https://arxiv.org/abs/2605.09165
[2605.09165] Sparse Layers are Critical to Scaling Looped Language Models
Looped language models repeat a set of transformer layers through depth, reducing memory costs and providing natural early-exit points at loop boundaries. However, looped models do not scale as favorably as standard transformers with unique layers. We compare standard and Mixture-of-Experts (MoE) transformers, with and...
Ryan Lee; Jacob Biloki; Edward J. Hu; Jonathan May
2026-05-09
2026
arXiv
arXiv
cs.LG
arxiv-api
2605.09165
2026-08-07T12:31:05