← ClaudeAtlas

artifacts-auditlisted

Audit a codebase to determine which canonical engineering artifacts (READMEs, ADRs, design docs, diagrams, runbooks, postmortems, ERDs, etc.) it should have on file, then propose a concrete generation + maintenance plan. Runs a 5-phase discovery → profile → audit → plan → deliverable flow and writes the result to `docs/artifacts-plan.md`. Plans only — does not generate the artifacts themselves and does not modify source code. Use when the user invokes `/artifacts-audit`, or asks to audit / inventory / plan a repo's documentation, diagrams, ADRs, runbooks, or other engineering artifacts.
ksdisch/claude-config · ★ 0 · AI & Automation · score 70
Install: claude install-skill ksdisch/claude-config
# Engineering artifacts audit Audit the current codebase against a canonical taxonomy of engineering artifacts (docs, diagrams, plans, ops material, knowledge) and produce a single planning document plus a chat summary. This skill **plans only** — generation is a follow-up session. The canonical taxonomy is in [reference/taxonomy.md](reference/taxonomy.md). Skim it before Phase 3. ## When this skill fires - User types `/artifacts-audit` - User asks something like: "audit this repo's docs", "what artifacts should this project have?", "plan engineering documentation", "what's missing from this codebase's documentation?", "inventory our ADRs / diagrams / runbooks" ## What this skill does NOT do - **Does not modify source code.** Hands off on `src/`, configs, CI, etc. - **Does not generate the artifacts themselves** (no READMEs, ADRs, diagrams written this session). That's the follow-up. - **Does not skip Phase 2.** The project profile + pause for confirmation is the load-bearing step. Don't shortcut it. - Creating `docs/artifacts-plan.md` and the `docs/` directory IS allowed — that's the deliverable. ## Phase 1 — Discovery Explore the repo systematically. Use the most efficient tool for each step (Bash for `git`/`find`, Glob/Grep for surveys, Read for files). If the repo is large or unfamiliar, dispatch the Explore subagent for the broader sweeps. Minimum work: 1. **Inventory the tree.** Top-level files + directories (3 levels deep), language breakdown, total file count