| name | managing-claude-code-meta |
| description | MUST be loaded when setting up, installing, migrating, reviewing, auditing, or checking CLAUDE.md files in projects. Covers installing the promode CLAUDE.md into new projects, migrating existing CLAUDE.md content to AGENT_ORIENTATION.md (progressive disclosure), and auditing projects for conformance. Invoke PROACTIVELY when user mentions CLAUDE.md, project setup, agent configuration, or code meta files. |
1. CLAUDE.md is for main agent behaviour CLAUDE.md defines the main agent's role: conversing with users, delegating to sub-agents, and following the promode methodology. It does NOT contain project-specific technical details — those belong in AGENT_ORIENTATION.md.
2. Sub-agents use phase-specific agents Claude Code sub-agents don't inherit CLAUDE.md. Promode provides phase-specific agents (implementer, reviewer, debugger) with the methodology baked in. Main agents handle brainstorming, planning, and orchestration directly, then delegate execution to the appropriate phase agent.
3. AGENT_ORIENTATION.md is the agent knowledge graph Each package/directory can have an AGENT_ORIENTATION.md with compact, token-efficient guidance for agents. This is distinct from README.md (which is for humans). Agents read these just-in-time when working in that area.
4. Tests are the documentation Long-lived markdown should cover architecture and principles only. Detailed behaviour documentation belongs in executable tests. If behaviour isn't tested, it's not guaranteed.
5. CLAUDE.md is standardised
The standard CLAUDE.md (standard/MAIN_AGENT_CLAUDE.md) should be copied exactly into projects. It is designed to work universally. All project-specific content belongs in AGENT_ORIENTATION.md.
- Install — Set up promode in a new project (no existing CLAUDE.md)
- Update — Update an existing promode installation to the latest version
- Migrate — Refactor a non-promode CLAUDE.md, moving content to AGENT_ORIENTATION.md
- Audit — Check if a project follows progressive disclosure principles
Wait for response before proceeding.
Intent-based routing:
- "set up CLAUDE.md", "add agent config", "install promode" → workflows/install.md
- "update promode", "get latest", "upgrade", "update CLAUDE.md" → workflows/update.md
- "CLAUDE.md is too big", "slim down", "refactor existing" → workflows/migrate.md
- "is this right?", "check conformance", "audit" → workflows/audit.md
Key distinction:
- Install = No promode yet, start fresh
- Update = Promode exists, bring to latest version and ensure all components present
- Migrate = Has CLAUDE.md but it's not promode (contains project-specific content)
- Audit = Read-only check, no modifications
After reading the workflow, follow it exactly.
Sub-agents: Main agents delegate execution to phase-specific agents (implementer, reviewer, debugger), which already know the methodology. Brainstorming, planning, and orchestration are done by the main agent.
Promode project structure:
project/
├── CLAUDE.md # Main agent behaviour (promode methodology)
├── KANBAN_BOARD.md # Project tracking across sessions
├── AGENT_ORIENTATION.md # Compact agent guidance (tools, patterns, gotchas)
├── .mcp.json # MCP server configuration
├── README.md # Human documentation (optional, for GitHub etc)
└── packages/
└── {package}/
└── AGENT_ORIENTATION.md # Package-specific agent guidance
| Workflow | Purpose |
|---|---|
| install.md | Install promode into new project |
| update.md | Update existing promode installation to latest version |
| migrate.md | Migrate non-promode CLAUDE.md content to AGENT_ORIENTATION.md |
| audit.md | Audit project for progressive disclosure conformance |
Required:
- CLAUDE.md — exact copy of
standard/MAIN_AGENT_CLAUDE.md - KANBAN_BOARD.md — project tracking across sessions
- AGENT_ORIENTATION.md — compact agent guidance at project root
- .mcp.json — MCP servers configured (context7, exa, grep_app)
Recommended:
- Package AGENT_ORIENTATION.md files for domain-specific context
- LSP configured for detected languages
- Tests document system behaviour, not markdown files
- README.md exists for humans (GitHub, etc) but is not agent-oriented