ai-workbench / templates
AGENTS.md0 starsChanged 47 days agoMITView on GitHub
1. Markdown is the state. Plans, decisions, status and learnings live here as versioned markdown. No second tracker; nothing that matters stays in a chat transcript. 2. AI at the edges, deterministic core. The agent drafts, triages, summarizes. Anything that must be reproducible is a script under scripts/. 3. The desk file is rewritten, never appended. STATUS.md states what is true now and who owes what. Anything longer than two lines moves to its own document, and the desk carries…
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- templates/CLAUDE.mdCLAUDE.md
- session-closeSkill
- session-openSkill
- workbench-scaffoldSkill
The file
Files
AGENTS.md — <project>
Purpose
<Two to four lines: what this repo is, who works in it, what it produces. State the one thing an agent would get wrong without being told.>
Rules
- Markdown is the state. Plans, decisions, status and learnings live here as versioned markdown. No second tracker; nothing that matters stays in a chat transcript.
- AI at the edges, deterministic core. The agent drafts, triages, summarizes.
Anything that must be reproducible is a script under
scripts/. - The desk file is rewritten, never appended.
STATUS.mdstates what is true now and who owes what. Anything longer than two lines moves to its own document, and the desk carries the link. - A decision that changes behaviour gets an entry in
DECISIONS.mdwith its why. A lesson that cost more than fifteen minutes becomes a rule here or a note there. - No silent drops. Anything unhandled, out of scope or uncertain is surfaced at session close, in the desk file or the hand-over.
- Caps. This file stays under 100 lines and
STATUS.mdunder 120. The lint enforces both; overflow moves to a linked document. - <Domain rules that would otherwise cause a real mistake, one line each. Earned by mistakes, not guessed at setup.>
Layout
STATUS.md— the desk file, read at every session open.DECISIONS.md— append-only decision log.scripts/—check(the whole CI) andworkbench_lint.py..claude/skills/—session-open,session-close.<Only the non-obvious folders. Never a file-by-file tour.>
Commands
scripts/check— lint plus tests; must pass before every push. Read-only.python3 scripts/workbench_lint.py— caps and links alone. Read-only.- <The few commands that matter, one line each; mark writes and anything that costs money.>
Working here
Open with session-open: read the log, the desk, the check and the tracker; brief;
propose one next step. Close with session-close: capture decisions and learnings, rewrite
STATUS.md, run scripts/check, commit on a branch, open a pull request that references
its issue. Issues are the only backlog (labels: bug, task, decision, blocked). Activate
the hook once per clone with git config core.hooksPath .githooks.
Context
<Business or domain facts an agent cannot derive from the tree: people, cadence, systems of record, what must never leave this repo. Under ten lines.>
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.

