ROADMAP.md

Roadmap

This roadmap keeps Awesome Loop Engineering focused on useful, verifiable work for an emerging practice.

Near Term

• Collect more direct Loop Engineering sources as the term stabilizes.
• Add real or anonymized gallery entries from practitioners running recurring agent loops.
• Grow the runnable loop directory beyond the test-repair reference loop, including scheduled-trigger variants per runtime.
• Add more translations for the introduction, mental model, Loop Contract, and contribution guide.
• Continue replacing weak or unstable links with primary sources, official docs, papers, and implementation-heavy write-ups.

Pattern Library

Completed reference patterns now cover PR babysitting, CI repair, docs drift, deploy verification, feedback clustering, dependency triage, evaluation regression, security review, cost-control, and bug hunting. Every pattern now ships a schema-validated loop contract in examples/.

Next pattern-library work should focus on variants backed by operational evidence:

• enterprise approval loops for permissioned agents;
• incident response loops with pager handoff and postmortem evidence;
• data-quality loops for repeated dataset or knowledge-base refresh;
• release-note loops that connect commits, issues, docs, and changelogs;
• model-routing loops that balance quality, latency, privacy, and cost.

Gallery

The gallery should grow from reference examples into public or anonymized case studies. Good entries should include:

• the runtime or agent tool used;
• trigger and intake source;
• verification gates;
• durable state artifact;
• budget and escalation rules;
• receipts or anonymized evidence;
• lessons learned after real use.

Quality And Governance

• Keep CI dependency-light and easy for contributors to run locally.
• Keep all resource annotations tied to recurring agent systems, not generic AI-agent interest.
• Keep public claims conservative: this repository is an early curated field guide, not a finished standard.
• Preserve clean owner-only commit identity for main.

Open Questions

• Which loop primitives become common across Codex, Claude Code, GitHub Agentic Workflows, and custom runtimes?
• What is the right schema shape for portable loop contracts?
• Which verification gates are strong enough for unattended or semi-attended loops?
• How should maintainers evaluate submitted real-world loop examples without exposing private data?