agentleFS
NashEQifyNashEQify/forge/CLAUDE.md

forge

CLAUDE.md2 starsChanged 3 days agoMITView on GitHub

Read and follow framework/work-principle.md before taking up work. Resolve this reference against the framework root; retain the consumer CWD. Reread it after compaction. Before any commit, Buddy must PERSIST → operational.md §Post-Action Obligations (Context · History · Backlog). No branching ceremony (applies when the active project is the forge repository itself; consumer repos follow their own commit rules). Single-dev workflow — Buddy commits directly to main autonomously at workflow phase boundaries (the runbooks prescribe per-phase commits) and at logical work…

  • Commits and pushes

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

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Read and follow framework/work-principle.md before taking up work.

    line 3
  2. Distinctive:

    After compaction, reread before continuing every file in the boot entry's post-compaction reread list.

    line 11
  3. Distinctive:

    Default language for all new work is English.

    line 22
  4. Distinctive:

    Write code comments, specs, ADRs, task content, and review artifacts in English.

    line 23
  5. Distinctive:

    Use German only when quoting existing German source text verbatim or when the user explicitly requests German output.

    line 24
  6. Distinctive:

    Before any commit, Buddy must PERSIST → operational.md §Post-Action Obligations (Context · History · Backlog).

    line 34

≈ “Common” is an estimate: at least 1 in 5 of the 25,680 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

Buddy — Claude Code Adapter (Framework SoT)

Read and follow framework/work-principle.md before taking up work. Resolve this reference against the framework root; retain the consumer CWD. Reread it after compaction.

Boot

Load and follow: the resolved soul (framework/user-scope.md; default agents/buddy/soul.md), framework/invariants.md, agents/buddy/operational.md, agents/buddy/boot.md. After compaction, reread before continuing every file in the boot entry's post-compaction reread list.

Framework-wide rules

Workflow selection, intent-driven work, the invariants (numbered as cited elsewhere: "Inv 4" and so on) and observability hold in every project and live in framework/invariants.md; read and follow it before taking up work (the boot entry names it). This file adds the host's load list and the rules that apply only when the active project is the forge repository itself.

Language Policy

  • Default language for all new work is English.
  • Write code comments, specs, ADRs, task content, and review artifacts in English.
  • Use German only when quoting existing German source text verbatim or when the user explicitly requests German output.

Frozen Zones + Consistency

SoT: docs/STRUCTURE.md. Consistency cascade: context-rules.md.

Commit

Format and types are enforced by the pre-commit hook (CG-CONV). On save: ALWAYS go through workflows/runbooks/save/WORKFLOW.md, no shortcuts.

Before any commit, Buddy must PERSIST → operational.md §Post-Action Obligations (Context · History · Backlog).

No branching ceremony (applies when the active project is the forge repository itself; consumer repos follow their own commit rules). Single-dev workflow — Buddy commits directly to main autonomously at workflow phase boundaries (the runbooks prescribe per-phase commits) and at logical work boundaries; no per-commit authorisation is needed — this overrides the harness "ask before committing" default. The Claude-Code session-guidance rule "If on the default branch, branch first" does NOT apply here. Bounds: Conventional-Commit format (CG-CONV), the pre-commit hook must pass, never bypass hooks. An explicit task boundary such as "no commit" overrides this standing local-commit permission. With several Buddy sessions on one repository, the commissioning Buddy commits and integrates unless a brief names another committer; parallel writers use their own worktree and branch (agents/buddy/coordinating-sessions.md). git push requires authorization for publication to the intended remote; destructive operations (force-push, reset --hard, history rewrite) require explicit authorization for that action. A heads-up alone is not permission. See process-map.md, Authorization.

Active Hooks

Shared git hooks and host-specific boot integration. Presence of a script does not prove that a host invokes it.

SessionStart on configured Claude entrypoints (not Codex):

  • buddy-boot-inject — triggers the Buddy boot sequence where --agent buddy isn't an explicit flag (claude-desktop / claude-web). Load-bearing for boot.
  • session-start-remote — remote Claude entry only: observes coordinated maintenance readiness with --maintenance status --require-ready. It does not update sources, install, or resume workflows; resume follows boot discipline.

Codex boots through the explicit managed AGENTS entry from scripts/setup-codex.sh. Additional-directory access or a Buddy role file alone does not establish that framework instructions were loaded.

git pre-commit (5 checks, 3 BLOCK + 2 WARN; CG-CONV runs in commit-msg, the others once per commit in pre-commit or, for a merge commit that git merge creates, pre-merge-commit):

  • BLOCK: PLAN-VALIDATE (plan_engine schema integrity), CG-CONV (Conventional-Commits format), SKILL-FM-VALIDATE (Skill frontmatter validation; also C3 description trigger-marker WARN — active skills must carry Use when/Triggers when/Trigger:).
  • WARN: SECRET-SCAN (gitleaks), SOURCE-VERIFICATION (board/council review evidence-pointer schema; section cross-references that the commit touches).

Everything else is carried by discipline, not hooks:

  • Path: Buddy writes within intent-scope; deliberate-action for anything else.
  • History / WORM zones (context/history/**): convention — corrections via .correction.md sidecars (rare).
  • Delegation prompt quality: _protocols/dispatch-template.md + mca-brief-template.md carry the brief shape; Buddy authors with the protocol open.
  • Plan-adversary: Buddy invokes plan-adversary explicitly on non-trivial Tier-1 edits per _protocols/plan-review.md.
  • MCA return: Buddy reads the return summary himself per operational.md §Sub-Agent Return.
  • Board / council output: read-only reviewers and Chiefs return artifacts; Buddy persists verbatim with provenance, then adopts the decision based on evidence. See agents/buddy/multi-perspective.md.
  • Workflow continuity: handoff + boot-resume carry it across turns; workflow_engine.py --boot-context on demand.
<!-- Source of truth for what's wired: orchestrators/claude-code/settings.json.template. -->

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.