agentleFS
sasuketoriisasuketorii/rev_harness_public/CLAUDE.md

rev_harness_public

CLAUDE.md0 starsChanged 3 months agoApache-2.0View on GitHub

This file is auto-read by both Claude Code and Cursor CLI. It is intentionally vendor-neutral and contains only shared RevHarness bootstrap instructions. Vendor-specific operational rules are kept out of this file so that one agent family does not accidentally follow another family's runtime contract. Read AGENTS.md first for repository-wide RevHarness invariants. Claude Code orchestrators must then read .claude/CLAUDE-LOCAL.md before operating, because Claude-specific wrapper, delegation, state, and role-switching rules live there. Cursor-specific rules live under .cursor/rules/. Shared operating rules live under…

The file

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

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

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

    Read AGENTS.md first for repository-wide RevHarness invariants.

    line 8
  2. Distinctive:

    Use the numbered read order in AGENTS.md §Read Order for shared, vendor-neutral decisions.

    line 42
  3. Very specific:

    Read docs/canonical-invariants.md for the canonical invariant set.

    line 51
  4. Very specific:

    When the current operator is Claude Code, read .claude/CLAUDE-LOCAL.md at session start before planning, delegating, editing, reviewing, or …

    line 62
  5. Very specific:

    If .claude/CLAUDE-LOCAL.md is missing or unreadable during a Claude Code orchestrator session, fail closed and report that the …

    line 75
  6. Distinctive:

    If Cursor-specific rules are absent, do not invent delegation, wrapper, or runtime behavior from this file.

    line 85
  7. Distinctive:

    Every agent must preserve required deterministic checks, command results, covered scope, and evidence pointers according to the active …

    line 111
  8. Very specific:

    Use .claude/tmp/<task>/ or .agent/active/ for task evidence unless the slice contract names another destination.

    line 116
  9. Distinctive:

    Do not write raw secrets, credentials, tokens, cookies, or unredacted sensitive payloads into prompts, logs, artifacts, tests, screenshots, …

    line 120
  10. Distinctive:

    If project identity is missing or malformed, stop and report the blocker instead of creating a new identity opportunistically.

    line 129
  11. Distinctive:

    Keep this file short and vendor-neutral.

    line 134

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 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

RevHarness Vendor-Neutral Bootstrap

This file is auto-read by both Claude Code and Cursor CLI. It is intentionally vendor-neutral and contains only shared RevHarness bootstrap instructions. Vendor-specific operational rules are kept out of this file so that one agent family does not accidentally follow another family's runtime contract.

Read AGENTS.md first for repository-wide RevHarness invariants. Claude Code orchestrators must then read .claude/CLAUDE-LOCAL.md before operating, because Claude-specific wrapper, delegation, state, and role-switching rules live there. Cursor-specific rules live under .cursor/rules/. Shared operating rules live under .agent_rules/ and are incorporated by reference via .agent_rules/RULES.md; read that literal path as part of the root contract.

Scope

This file is the root bootstrap for agents that auto-read CLAUDE.md. It points to the authoritative instruction locations and defines only the shared read order for this repository.

It is not the source of truth for Claude-only orchestration, Codex-only runtime policy, Cursor project rules, reviewer verdicts, or acceptance closure.

Shared Sources

  • AGENTS.md: vendor-neutral RevHarness invariants for every agent family.
  • .agent_rules/RULES.md: shared-rule index and incorporation point for the shared modules under .agent_rules/.
  • docs/canonical-invariants.md: on-disk canonical anchor for the invariant set; read this for the full current invariant set rather than rediscovering rows from root summaries.
  • .claude/CLAUDE-LOCAL.md: Claude Code orchestrator-specific operating rules.
  • .cursor/rules/: Cursor-specific project rules and attachment behavior.
  • docs/roles/*.md: canonical role definitions.
  • docs/manual/verification-truth-matrix.md: acceptance and evidence authority.
  • .agent/PROJECT_CONTEXT.md: project-specific context.
  • .shared/project_id: immutable project identity.

Truth Read Order

Use the numbered read order in AGENTS.md §Read Order for shared, vendor-neutral decisions. This file intentionally does not restate that list.

Vendor-specific rules may narrow how an agent acts, but they do not override the truth matrix's deterministic acceptance requirements.

For the shared release-invariant gates, the following references apply regardless of operating agent family:

  • Read docs/canonical-invariants.md for the canonical invariant set.
  • For phase advance, I-12 (smoke-gated dual-LGTM) governs the state-transition-guard --require-lgtm-final gate. The mechanism's normative prose is owned by docs/canonical-invariants.md §I-12 — read it there; this file does not restate the smoke / lgtm_stage / phase-advance details.
  • Vendor-family-specific delegation, wrapper, or role-switching guidance for these release gates belongs in .claude/CLAUDE-LOCAL.md (Claude Code) or under .cursor/rules/ (Cursor) — not in this file.

Claude Bootstrap

When the current operator is Claude Code, read .claude/CLAUDE-LOCAL.md at session start before planning, delegating, editing, reviewing, or reporting. The root file is intentionally too small to carry Claude orchestration details.

In addition, the session-start rule (invoke the orchestrator-bootstrap skill before the first substantive action) has a single canonical home: AGENTS.md §Session Start (Orchestrator session). Read that section for the authoritative rule; this file does not restate it. The only Claude-specific delta is the invocation mechanism: a Claude Code orchestrator invokes orchestrator-bootstrap via the Skill tool (Codex uses its native skill mechanism). The session-start rule itself is identical across families. See the Authority Map in docs/canonical-invariants.md §Authority Map.

If .claude/CLAUDE-LOCAL.md is missing or unreadable during a Claude Code orchestrator session, fail closed and report that the Claude-specific operating contract is unavailable.

Cursor Bootstrap

Cursor agents should treat this file as shared context only. Cursor-specific rules, including rule attachment and mode-specific behavior, belong in .cursor/rules/ and must not be inferred from Claude or Codex documents.

If Cursor-specific rules are absent, do not invent delegation, wrapper, or runtime behavior from this file. Follow the user instruction, AGENTS.md, and the truth matrix.

Codex Bootstrap

Codex agents should treat this file as shared context only. Codex runtime configuration and native subagent presets live under .codex/; external caller-facing execution contracts are documented with the canonical wrappers in scripts/.

This bootstrap does not redefine Codex model policy, sandbox policy, or reviewer authority.

Shared mirror for dual-native orchestration: Claude top-level sessions use Claude-native subagents, Codex top-level sessions use Codex native subagents, and cross-family handoffs use artifact packets. The shared wrapper role map and compatibility shim mapping are owned by .agent_rules/shared-delegation.md.

Acceptance

docs/manual/verification-truth-matrix.md is the authority for acceptance, worker outcome, evidence placement, reviewer verdict validity, and completion language. A response, wrapper invocation, or local convention is not enough to claim acceptance.

Every agent must preserve required deterministic checks, command results, covered scope, and evidence pointers according to the active slice contract.

Evidence And Secrets

Use .claude/tmp/<task>/ or .agent/active/ for task evidence unless the slice contract names another destination. Keep artifacts traceable enough for a later reviewer to replay the decision.

Do not write raw secrets, credentials, tokens, cookies, or unredacted sensitive payloads into prompts, logs, artifacts, tests, screenshots, or reports. Prefer redacted previews and record when raw output was intentionally withheld.

Project Identity

.shared/project_id is immutable project identity. Do not regenerate or edit it during routine instruction, wrapper, or rule migrations.

If project identity is missing or malformed, stop and report the blocker instead of creating a new identity opportunistically.

Maintenance Rule

Keep this file short and vendor-neutral. If an update adds agent-family-specific commands, wrapper details, role-switching obligations, state schemas, or delegation mechanics, move that content to the vendor-specific location and leave only a pointer here.

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.