§ reference · skill

annotate-inline

Documents a project's architecture for AI assistants by writing compact `CLAUDE.md` files inline next to the source they describe.

arguments target-directory

§ 01 · purpose

Purpose

Same end goal as annotate-guidance — onboards AI agents to a brownfield codebase via per-directory architecture notes — but places CLAUDE.md files next to the source they describe instead of in a shadow tree. Choose this when teams prefer guidance to live beside the code.

§ 02 · when to use

When to use it

  • The project should keep architecture docs adjacent to source.
  • You need first-time AI onboarding for a codebase that already uses inline CLAUDE.md conventions.
  • Pick annotate-guidance instead for shadow-tree (.rpiv/guidance/) layout.
  • Pick migrate-to-guidance when existing inline files need to move into the shadow tree.

§ 03 · inputs

Inputs

Name Required Source
target-directory no CLI argument or current working directory
existing CLAUDE.md / READMEs no Paths mentioned inline
existing CLAUDE.md / READMEs
Read fully before agents dispatch.

§ 04 · outputs

Outputs

Artifact Path Format
Root project guidance CLAUDE.md (project root) markdown
Per-layer guidance <layer>/CLAUDE.md (next to source) markdown

§ 05 · key steps

Key steps

  1. Read directly-mentioned files first Why: Existing docs lock in team-level decisions before any subagent runs, so later passes don't contradict what's already documented.
  2. Pass 1 — map tree and detect architecture (parallel) Why: One locator agent maps the tree; one identifies the architecture from folder shape and manifest files. Running both concurrently halves discovery time.
  3. Apply guidance-depth rules; propose targets Why: A two-pass selection (top-level layers, then opt-in decomposition) keeps the target list grounded in folder shape and is shown to the user for confirmation before any analysis spend.
  4. Pass 2 — analyzer + pattern-finder per target (parallel) Why: Per-target codebase-analyzer answers "what is this for"; codebase-pattern-finder extracts idiomatic code shapes. Both run concurrently for depth per token.
  5. Developer checkpoint on findings Why: One question at a time surfaces deprecated patterns, undocumented conventions, and migrations-in-progress that code reading alone cannot reveal.
  6. Batch-write CLAUDE.md files Why: Files are written in a single emission pass after the checkpoint, so the inline-guidance set lands internally consistent.

§ 06 · related skills

downstream researchdesignblueprint