ai-development-patterns
AGENTS.md674 starsChanged 4 months agoMITView on GitHub
Tasks live in .tasks/backlog.jsonl — an append-only JSONL file, one task per line, git-tracked. There is no daemon and no database: the file is the record and git is the history. Do NOT use TodoWrite, TaskCreate, markdown TODO lists, or bd/beads — this repo migrated off beads on 2026-08-17. A task's state is its LAST line. The file is append-only, so grep '"status": *"open"' also matches tasks that were later closed, over-reporting the open set. Always fold last-line-wins: Close a…
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- evaluate-pattern-adoptionSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 1 is common, 4 are distinctive, and 2 are very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 7
Do NOT use TodoWrite, TaskCreate, markdown TODO lists, or bd/beads — this repo migrated off beads on 2026-08-17.
- Very specific: line 23
Run model-backed pattern-adoption research only through the repository-local $evaluate-pattern-adoption skill in an interactive, signed-in …
- Common: line 26
Never run it in GitHub Actions, expose evaluator API keys to the repository, or bypass either the exact plan approval or the separate …
- Distinctive: line 48
Verify: - All changes committed AND pushed
- Distinctive: line 53
NEVER stop before pushing - that leaves work stranded locally
- Distinctive: line 54
NEVER say "ready to push when you are" - YOU must push
- Distinctive: line 55
If push fails, resolve and retry until it succeeds
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent Instructions
Task Tracking
Tasks live in .tasks/backlog.jsonl — an append-only JSONL file, one task per line, git-tracked. There is no daemon and no database: the file is the record and git is the history.
Do NOT use TodoWrite, TaskCreate, markdown TODO lists, or bd/beads — this repo migrated off beads on 2026-08-17.
A task's state is its LAST line. The file is append-only, so grep '"status": *"open"' also matches tasks that were later closed, over-reporting the open set. Always fold last-line-wins:
python3 -c "import json;last={};[last.__setitem__(t['id'],t) for t in map(json.loads,filter(str.strip,open('.tasks/backlog.jsonl')))];print('\n'.join(f\"{i} {t['title'][:68]}\" for i,t in last.items() if t['status']=='open'))"Close a task by APPENDING a new line with status set to closed; never edit or delete an existing line. Work discovered while doing another task gets its own line with discovered_from set to the parent id.
Schema: {id, title, status, blocks[], depends_on[], spec_ref, created, priority, description, discovered_from?} — status is open or closed; priority is 1 (most urgent) to 4; a task is ready when every id in depends_on is closed. Issue ids keep their original beads form so existing commit messages still resolve.
The full beads history is archived verbatim at .tasks/beads-archive.jsonl.
Pattern Adoption Evaluation
Run model-backed pattern-adoption research only through the repository-local
$evaluate-pattern-adoption skill in an interactive, signed-in local evaluator client
(Codex today).
Never run it in GitHub Actions, expose evaluator API keys to the repository, or bypass either the
exact plan approval or the separate draft-PR publication approval. Use no more than three
read-only research subagents at once, keep the root task as the sole writer, and use a fresh
read-only verifier for every batch. GitHub Actions is limited to deterministic validation and
read-only link/content checks for this capability.
Landing the Plane (Session Completion)
When ending a work session, you MUST complete ALL steps below. Work is NOT complete until git push succeeds.
MANDATORY WORKFLOW:
- File tasks for remaining work - Append tasks to
.tasks/backlog.jsonlfor anything that needs follow-up - Run quality gates (if code changed) - Tests, linters, builds
- Update task status - Append
status=closedlines for finished work - PUSH TO REMOTE - This is MANDATORY:
git pull --rebase git push git status # MUST show "up to date with origin" - Clean up - Clear stashes, prune remote branches
- Verify - All changes committed AND pushed
- Hand off - Provide context for next session
CRITICAL RULES:
- Work is NOT complete until
git pushsucceeds - NEVER stop before pushing - that leaves work stranded locally
- NEVER say "ready to push when you are" - YOU must push
- If push fails, resolve and retry until it succeeds
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.

