context-init

context-init

Read-only codebase exploration mode. Build a mental model of the project sufficient to contribute meaningfully — without modifying anything.

Guardrails

• Read-only. No Edit, Write, NotebookEdit, or state-mutating commands (commits, pushes, installs, deletes, git reset, etc.).
• Breadth before depth. Get a shape-of-the-codebase overview first; drill into a specific area only if the user asks.
• Parallel exploration is encouraged for large repos — spawn subagents for independent axes (stack, git state, infra, tests).

What to figure out

1. What is this? — Purpose, domain, intended users.
   • Read: README*, CLAUDE.md, AGENTS.md, docs/, and the root-level project manifest.
2. Tech stack — Languages, frameworks, dependencies, build/test/lint tooling.
   • Inspect whichever exist: package.json, pyproject.toml / uv.lock, Cargo.toml, go.mod, Gemfile, pom.xml, build.gradle, Dockerfile, .tool-versions, mise.toml.
3. Architecture — Directory structure, entry points, key abstractions, data flow, layering (MVC, hexagonal, monorepo, plugin-based, etc.).
4. Current state — git status, current branch, git log --oneline -20, open PRs/issues (gh pr list, gh issue list if gh is available and the repo is on GitHub).
5. Active work — TODO / FIXME / XXX / HACK markers (capped, don't flood), recent churn: git log --oneline --since='2 weeks ago' --name-only.
6. Infrastructure — Deployment configs: Dockerfile, docker-compose*, .github/workflows/, terraform/, k8s/, cloud manifests. Identify target environments if discoverable.
7. Testing — Test runner, where tests live, any integration-test subdirs. Don't run them.

Skip any section where no evidence exists — don't pad with speculation.

Output

Produce this report. Keep each section to a few lines unless the user asked for depth.

## Project
[What it is, who it's for — 2 sentences max]

## Stack
[Languages, frameworks, key deps, build/test/lint tools]

## Architecture
[Directory layout, key components, data flow, notable patterns]

## Current State
[Branch, uncommitted changes, recent activity, open PRs/issues]

## Active Work
[What's in progress, what's stale, what needs attention]

## Infrastructure
[Deploy target, CI/CD, cloud services — if applicable]

## Key Files
[Entry points, configs, files worth reading first — absolute paths]

$ARGUMENTS