agentleFS
Sign inStart free
AI-Driven-SchoolAI-Driven-School/aiki/CLAUDE.md

aiki

CLAUDE.md0 starsChanged 9 months agoMITView on GitHub

Always record architecture, technology choices, and design decisions: Before ending a session: - Ask "Update CLAUDE.md with today's work" - Or manually update the Work History section Latest important decisions (see docs/decisions/ for details):

The file

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

9 rules. About 0 are common, 6 are distinctive, and 3 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.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    NEVER implement source code directly: when the task matches Codex delegation criteria:

    line 7
  2. Distinctive:

    Test file creation → MUST delegate to Codex

    line 10
  3. Very specific:

    Use /requirements and /spec to create them first

    line 17
  4. Distinctive:

    Always record architecture, technology choices, and design decisions:

    line 60
  5. Distinctive:

    Before ending a session:

    line 68
  6. Distinctive:

    Ask "Update CLAUDE.md with today's work"

    line 69
  7. Very specific:

    Always write requirements in docs/requirements/ before adding new features

    line 74
  8. Distinctive:

    Default policy: delegate implementation to Codex

    line 75
  9. Very specific:

    When user says "record this decision", save to docs/decisions/

    line 77

≈ “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 Workflow Rules

Rule Enforcement (MANDATORY — DO NOT IGNORE)

HARD RULES — These are NOT suggestions. Violations will be BLOCKED by hooks.

  1. NEVER implement source code directly when the task matches Codex delegation criteria:
    • 3+ new source files → MUST delegate to Codex via /implement
    • 5+ source file edits → MUST delegate to Codex via /implement
    • Test file creation → MUST delegate to Codex
    • Boilerplate/repetitive code → MUST delegate to Codex
    • Enforced by: enforce-delegation.sh (PreToolUse hook, exit code 2 = BLOCK)
  2. NEVER start implementation without design artifacts:
    • docs/requirements/{feature}.md MUST exist before writing source code
    • docs/specs/{feature}.md MUST exist before writing source code
    • Use /requirements and /spec to create them first
    • Enforced by: agent-router.sh (design-first warning)
  3. Claude's allowed domains (NOT blocked):
    • Documentation (.md files)
    • Configuration (.json, .yaml, .toml)
    • Design artifacts (docs/, .claude/, skills/, scripts/)
    • Requirements, specs, reviews, API design
    • Code review and architecture decisions
    • Hook/script infrastructure
  4. Override: User can set export FORCE_CLAUDE=1 to bypass all blocks.
  5. Session counters: Reset with rm .claude/.session_edit_count .claude/.session_new_files .claude/.session_edit_log

Command System

CommandAIPurpose
/project <feature>AllComplete flow: design -> implementation -> deploy
/requirementsClaudeRequirements definition
/specClaudeUI specifications
/implementCodexImplementation
/reviewClaudeReview
/checkpointingClaudeSave session state

Claude Code Orchestra

Auto-collaboration suggestions (Hooks)

KeywordSuggested AIExample
implement, createCodex"Implement authentication"
testCodex"Write unit tests"
research, analyzeGemini"Compare React state management"
compare, libraryGemini"Evaluate auth libraries"
trend, buzz, viralGrok"What's trending in React community"
X search, realtimeGrok"Latest developer reactions to Next.js 15"

Working Rules

1. Record important decisions

Always record architecture, technology choices, and design decisions:

docs/decisions/YYYY-MM-DD-title.md

2. End of session

Before ending a session:

  • Ask "Update CLAUDE.md with today's work"
  • Or manually update the Work History section

Notes for Claude

  • Always write requirements in docs/requirements/ before adding new features
  • Default policy: delegate implementation to Codex
  • Save benchmark results in benchmarks/
  • When user says "record this decision", save to docs/decisions/
  • Record design principles in .claude/docs/DESIGN.md
  • Gemini research results → .claude/docs/research/
  • Grok research results → .claude/docs/research/ (prefixed with grok-)

Important Decisions

Latest important decisions (see docs/decisions/ for details):

  • 2026-03-13: Integrated rulesync for unified AI rule management
  • 2026-02-14: Added Grok (xAI) as 4th AI - real-time trend & X search specialist
  • 2026-02-03: Integrated Claude Code Orchestra (Hooks + Rules + Knowledge Base + Checkpointing)

Work History

2026-05-08

  • Added Multi-Agent Debate case study (Solver / Proposer / Checker pattern)
    • docs/examples/multi-agent-debate-case-study.md — 1ジョブで法律解釈の事実誤認を検出した実例
    • README に動線追加（"Real-world Case" subsection under Case Study）
    • 検証手法: Solver(Claude) → Proposer(Gemini) → Checker(Claude+WebSearch)
    • 参考研究: MARCH (2026), Microsoft CORE — false positive -25.8%
    • 将来の aiki debate <claim-file> サブコマンド化候補として記録

2026-03-13

  • Integrated rulesync for unified AI rules management
    • .rulesync/rules/ - Single source of truth for all AI rules
    • rulesync generate auto-generates CLAUDE.md, AGENTS.md, GEMINI.md
    • rulesync install for declarative skill management
    • Added rulesync setup to install-fullstack.sh

2026-02-14

  • Added Grok (xAI API) integration as 4th AI

2026-02-03

  • Integrated Claude Code Orchestra

2025-02-03

  • Created CLAUDE.md, established memory persistence rules

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