claude-code-loop-patterns
AGENTS.md2 starsChanged 4 months agoMITView on GitHub
Complete the requested work through verification. Make reasonable local decisions and ask only when a missing answer changes the outcome; continue independent work while waiting. Preserve user edits and carry outstanding requests across interruptions. Use the tools and model actually available in this session. Skills supply task guidance within the user's scope and existing authorization. Keep simple work local. When delegation is authorized and useful, give workers bounded ownership, keep at most three active across the whole task, and inspect…
In this repository3 other files it gives its agents.
- loop-failure-triageSkill
- read-before-editSkill
- verify-before-doneSkill
The file
Files
Claude Code loop patterns
Working agreement
Complete the requested work through verification. Make reasonable local decisions and ask only when a missing answer changes the outcome; continue independent work while waiting. Preserve user edits and carry outstanding requests across interruptions.
Use the tools and model actually available in this session. Skills supply task guidance within the user's scope and existing authorization. Keep simple work local. When delegation is authorized and useful, give workers bounded ownership, keep at most three active across the whole task, and inspect and integrate their results.
Run checks relevant to the change plus required repository gates. Broaden testing only for new changes, failures, or unresolved risk. Report the result, evidence, and actual limits in concise plain prose. Commits, pushes, publishing, messages, credential changes, and destructive actions need authorization covering the action; do not request it again when already given. Do not add agent or model attribution.
Repository context
This repository contains Python reference implementations and Claude Code workflow samples. Editing a sample does not activate its Stop hook or make its verification ceremony the workflow for this session. Keep the Claude runtime contracts and historical research claims intact unless the task changes them.
Each numbered example should remain usable on its own with the Python standard library. Use synthetic inputs and temporary output directories; do not inspect personal session logs for routine verification.
Verification
Python 3.11+; install development tools with uv sync --locked --dev when needed. Run relevant uv run pytest cases during development. The CI gates include uv run ruff check ., uv run ruff format --check ., the targeted mypy command in .github/workflows/ci.yml, and uv run pytest --cov=. --cov-report=term-missing. Apply them to relevant code changes; prose edits need reference checks.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.