← ClaudeAtlas

censuslisted

Discover design decisions that exist in code but have no DR, and produce a DR promotion candidate list ranked by impact and reversibility. Pairs with adrift, which scans existing DRs for drift against code.
thkt/dotclaude · ★ 12 · AI & Automation · score 78
Install: claude install-skill thkt/dotclaude
# /census - DR gap audit ## Input `$ARGUMENTS` is an optional path naming the audit scope. What gets collected is set by the Phase 1 table. When scoped to a path, record the target in the report Summary's Scope row. ## Criteria Every criterion lives in ${CLAUDE_SKILL_DIR}/references/decision-criteria.md. That file holds impact / reversibility, the incomplete-contract definition, the DR-worth rule of thumb, and the challenge angles. ## Phase 1: Collect List source by running ${CLAUDE_SKILL_DIR}/scripts/list-source-files.py with python3. Scan for docs using the file patterns in ${CLAUDE_SKILL_DIR}/references/detection-targets.md. When source exceeds the guideline of 20, confirm narrowing via AskUserQuestion before the Phase 2 reviewer fan-out. Options are a subdirectory, top-N, or a specific module. Where each stream looks is set by the table below. | $ARGUMENTS | source | doc | | ----------- | --------------- | --------------------- | | none | repository root | top-level and `docs/` | | a directory | that path | that subtree | | a file | that file alone | nothing | ## Phase 2: Mine Record findings under the table columns in ${CLAUDE_SKILL_DIR}/templates/report-template.md, Source File Decisions for source and Prose Document Decisions for docs. Evidence is a comment, a name, a module-doc, or a commit, and a commit-derived one reads `commit <sha>`. ### Step 1: From source Two streams feed this step,