init-claude

Installation
SKILL.md

Init CLAUDE.md

Write CLAUDE.md: the operating manual an agent reads before touching the repo. It is not a doc like the others. The pipeline docs are the source of truth and the place to read in depth; CLAUDE.md is the dense distillation of the load-bearing rules plus pointers to those docs, written so an agent with limited context still gets the invariants right.

This runs at the end of the documentation pipeline, just before build: the technical specs, coding standard, task breakdown, and deployment plan already exist, and this synthesizes them. Three phases: read everything, grill the user on what the docs leave open, then write the manual. Because every agent reads this file before touching the repo, a wrong or guessed invariant here misleads every future build, so resolve the gaps before writing, never invent.

1. Read the pipeline docs

CLAUDE.md is a synthesis, so read its sources first and pull the real values, never invent them:

  • docs/technical-specs/: tech stack and versions (04), repo layout (03), module/service boundaries (05), data model and state machines (06), env config (11).
  • docs/coding-standard/ (or CODING_STANDARD.md in an older project): the architecture laws, naming, type-safety, error-handling, and testing rules to restate tersely.
  • docs/task-breakdown/ (or docs/TASK_BREAKDOWN.md in an older project): the work model (pre-assigned vs self-pick, scaffold model, reviewer).
  • docs/deployment-plan/ (or DEPLOYMENT_PLAN.md in an older project): environments, deploy triggers, the verification/build commands.
  • docs/GLOSSARY.md (or GLOSSARY.md at the root in an older project) and docs/business/: domain terms, roles, language policy.
  • Any existing CLAUDE.md to update in place rather than overwrite.

1b. Grill on the gaps

Installs
29
GitHub Stars
9
First Seen
Sep 30, 2026