| name | init-context |
| description | Build a complete mental map of a project through phased exploration so you can work effectively without rediscovering things. Use when the user wants to: (1) Onboard onto a new or unfamiliar codebase, (2) Get a full project overview or briefing, (3) Understand project architecture, structure, and open work, (4) Initialize context at the start of a deep work session, (5) Explore and summarize a repository end-to-end. Triggers on requests like "explore this repo", "get familiar with this project", "give me a project overview", "onboard me", "init context", or any request to deeply understand a codebase before working on it. |
Build a complete mental map of this project so you can work effectively without rediscovering things. Follow these phases:
Phase 1: Direct Reading (your own context)
Find and read key documentation files directly — do NOT delegate these, you need them in YOUR context:
- Look for root-level markdown files: README, STATUS, CHANGELOG, CONTRIBUTING, ARCHITECTURE, ROADMAP, REFERENCE, WORKLOG, and any project-specific docs (CLAUDE.md, AGENTS.md, etc.)
- Look for a CLAUDE.md or similar agent instruction file
- Check for a
.claude/ directory with project config or memory files
- If the codebase is large - use a sub-agent with a
layered-summary skill for the code sources folder to generate (or update, if exists) the AGENTS.md summaries
Phase 2: Sub-agent Exploration (parallel)
Based on what you learned in Phase 1, spawn sub-agents to explore the major areas of the codebase. Each agent should read all documentation files in its area, skim key source files, and return a structured summary of: what's there, how it works, key decisions, and where to look for what.
Typical areas to explore (adapt based on actual project structure):
- Each major source directory — architecture, key modules, entry points, important types/interfaces
- Tests & CI — test structure, CI config, how to run tests
- Config & Infrastructure — deployment configs, environment setup, docker files, scripts
- Documentation subdirectories — any docs/, wiki/, knowledge-base/, research/, or analysis/ folders
- GitHub/GitLab history — previous commits list, closed PRs, closed Issues (using
gh if available)
- GitHub/GitLab current state — open issues, open PRs, recent commits and activity (
gh issue list, gh pr list if available)
Adjust the number and focus of sub-agents based on project size. Small project = 2-3 agents. Large monorepo = 5-6 agents, iterative.
Phase 3: Fill Gaps
After receiving all sub-agent summaries, identify gaps in your understanding:
- Are there key source files referenced in docs that you haven't seen?
- Is the data flow / request lifecycle clear end-to-end?
- Do you understand how to build, test, and run the project?
- Is there anything that you're still missing and need to see?
- Spawn additional sub-agents or read files yourself as needed
- Check recent git log (
git log --oneline -20) for context on latest changes
Phase 4: Synthesize
Provide a concise project briefing covering:
- Project overview: What is this? What problem does it solve?
- Current state: What phase/version? What works? What's in progress?
- Architecture map: How the major pieces fit together, data flow, key abstractions
- Open work: Issues, PRs, blockers, next steps
- Key numbers: Any important metrics, benchmarks, or thresholds
- Developer guide: How to build, test, run. Key commands.
- Where to find things: Quick reference for common lookups (config, entry points, types, tests)
The goal: after this runs, you should have FULL project context and never need to ask "where is X?" or "what does Y do?" for the common cases.
