agentleFS
Sign inStart free
galimbagalimba/agentic-builder-forge/CLAUDE.md

agentic-builder-forge

CLAUDE.md4 starsChanged 4 months agoApache-2.0View on GitHub

The standing rules above (imported from AGENTS.md) are the operating contract. Under Claude Code they are not advisory: they are enforced by the deterministic hooks in .claude/hooks/, wired via .claude/settings.json. A PreToolUse hook returning deny blocks the tool call even under --dangerously-skip-permissions — that is what makes the floor a floor, not a suggestion. This repository is infrastructure — a build harness — not a documentation or knowledge vault. The canonical completion gate is tests/run-all.sh: it discovers every test:* script…

  • Deletes or force-pushes

The file

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

3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    If you keep a sibling knowledge repo (e.g.

    line 13
  2. Very specific:

    When a Stop gate blocks, the reason it prints is your next unit of work.

    line 31
  3. Very specific:

    Run bash .forge/scripts/init.sh, then bash .forge/scripts/doctor.sh --post-init.

    line 44

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.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
CLAUDE.md

CLAUDE.md — agentic-builder-forge

@AGENTS.md

The standing rules above (imported from AGENTS.md) are the operating contract. Under Claude Code they are not advisory: they are enforced by the deterministic hooks in .claude/hooks/, wired via .claude/settings.json. A PreToolUse hook returning deny blocks the tool call even under --dangerously-skip-permissions — that is what makes the floor a floor, not a suggestion. This repository is infrastructure — a build harness — not a documentation or knowledge vault.

  • The harness contract is SPEC.md.
  • How to run a task and the done-condition are in README.md; the operating guide is in docs/.
  • If you keep a sibling knowledge repo (e.g. ../my-vault), it is read-only to the Forge.

The enforcement floor (Claude Code hooks)

HookEventWhat it enforces
pre-tool-use-deny.shPreToolUseHard deny for destructive command shapes: rm -rf outside the sandbox, force-push, push/commit to {{DEFAULT_BRANCH}}, --no-verify, secret-shaped literals, raw writes into .beads/**, and writes to the enforcement files themselves.
post-tool-use-format.shPostToolUseAuto-formats and lints every Write/Edit, using the commands in harness/targets.config.
stop-gate-tests.shStopBlocks ending a build turn while tests/run-all.sh is red; the failure is fed back into the loop.
stop-gate-intake.shStopBlocks ending an intake session before the spec has passed the clarify loop and been human-ratified.
pre-tool-use-clarify-gate.shPreToolUseEnforces the intake clarify/restate budgets (harness/intake.config) before spec writes.
session-start-witness.shSessionStartSelf-mints the floor's integrity baseline at session start (no committed hash); tampering after that point is detected.

The canonical completion gate is tests/run-all.sh: it discovers every test:* script in package.json, reports PASS / SKIP(75) / FAIL, and fails closed.

Stop-gate behavior

When a Stop gate blocks, the reason it prints is your next unit of work. Fix the red tests (or complete the intake step) and finish again. Never try to end the session by editing .claude/hooks/** or .claude/settings.json, weakening a gate, or blanking a test:* script — the deny hook blocks writes to the enforcement files, and gutting a check to go green is the cardinal failure. If a gate seems genuinely wrong, stop and show the human.

Customization checklist

  1. harness/targets.config — build/test/lint/format commands per target type (typescript default; python and static examples).
  2. harness/reviewers.config — review backends (ollama | claude-fresh | codex); shipped model names are examples.
  3. harness/repos.config — copy from harness/repos.config.example (gitignored target-repo map).
  4. Run bash .forge/scripts/init.sh, then bash .forge/scripts/doctor.sh --post-init.

Full reference: docs/configuration.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