agentleFS
Sign inStart free
AgriciDanielAgriciDaniel/agentic-habits/AGENTS.md

agentic-habits

AGENTS.md2 starsChanged 3 months agoMITView on GitHub

The canonical entrypoint for the package is skills/habits/SKILL.md. This file exists for runtimes that load a project-level AGENTS.md, and it governs work on this repository, not the behavior the package installs. This is a Claude Code package. Three of its four mechanisms are Claude Code only. Do not describe them as portable, here or in anything you write. Every measurement in skills/habits/references/verification.md was made on Claude Code 2.1.251. None of it is evidence about another harness. The budget rule carries…

  • 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 31,774 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 repositoryOne other file it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every measurement in skills/habits/references/verification.md was made on Claude Code 2.1.251.

    line 22
  2. Very specific:

    Every claim about how Claude Code behaves, or about measured model behavior, is tested with a recorded method, quoted with its source, or …

    line 38
  3. Very specific:

    Never edit a check= to make a failure disappear.

    line 41
  4. Distinctive:

    No em dashes, no en dashes.

    line 47
  5. Very specific:

    Never write to a real ~/.claude while working here.

    line 48

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Agentic habits: agent instructions

The canonical entrypoint for the package is skills/habits/SKILL.md. This file exists for runtimes that load a project-level AGENTS.md, and it governs work on this repository, not the behavior the package installs.

What ports to another harness, and what does not

This is a Claude Code package. Three of its four mechanisms are Claude Code only. Do not describe them as portable, here or in anything you write.

PartPortsDetail
The card format: When, Do, CheckyesPlain markdown. Paste into AGENTS.md, GEMINI.md, .cursor/rules/*.mdc, or .github/copilot-instructions.md.
Three tiers, the repair ladder, budgets, the 12 anti-habits, the judging methodyesProse discipline. Nothing harness-specific.
The <!-- habit: ... --> bookkeepingnoStripping is a measured Claude Code behavior, tested against a control on 2.1.251. Elsewhere assume the comment is in the prompt, visible to the model and paid for in tokens. Strip it before porting.
.claude/rules/ and paths: scopingnoCodex and Gemini load their instruction file whole, every session. Cursor's globs: is the nearest analogue and is untested here. Porting collapses four scopes into one, which removes the instrument this package calls the whole mechanism.
skills/habits/assets/gates/completion-gate.shnoReads a Claude Code Stop payload and a Claude Code JSONL transcript, and blocks with exit 2. Every one of those is Claude Code specific. The gated tier does not exist elsewhere.
agents/habit-judge.mdpartlyThe method ports. tools: Read, Grep, Glob does not, and that restriction is why the verdict means anything. Elsewhere the judge is a prompt asking not to edit, which is a hint. That distinction is what this package exists to make.
/habitsnoA Claude Code skill invocation.

Every measurement in skills/habits/references/verification.md was made on Claude Code 2.1.251. None of it is evidence about another harness. The budget rule carries elsewhere; its justification may not, and the alternative justification, a hard document-size cap in another harness, is filed as unverified.

Read order

  1. skills/habits/SKILL.md
  2. skills/habits/references/verification.md
  3. skills/habits/references/methodology.md
  4. CONTRIBUTING.md
  5. SECURITY.md

Operating rules for work on this repository

  • Every claim about how Claude Code behaves, or about measured model behavior, is tested with a recorded method, quoted with its source, or named as neither, and filed in verification.md or evidence.md. Nothing else goes in.
  • Never edit a check= to make a failure disappear. Sharpening a trigger after a miss is the repair ladder. Sharpening the grader is tampering.
  • A card changes in both places or neither: references/starter-pack.md and assets/starter/. CI compares the cards and the checks verbatim.
  • Counts stated in prose are claims about this repository. Recount before changing one. Never copy a count forward from an older file. CI asserts them.
  • No em dashes, no en dashes. CI fails on one.
  • Never write to a real ~/.claude while working here. Test installs with HABITS_INSTALL_HOME="$(mktemp -d)".
  • The stated tier stays markdown. Shell lives in skills/habits/assets/gates/ and .github/, nowhere else.
  • List the exact commands and outcomes. Do not describe skipped, offline-only, or blocked checks as passes.

Verification

bash .github/checks.sh
bash .github/test-gate.sh
bash .github/test-install.sh
bash .github/live-checks/gate-replay.sh
claude plugin validate . --strict

bash .github/live-checks/rules-canary.sh needs the claude CLI and spends model quota. Run it before a release and re-stamp the version at the top of verification.md.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free