← ClaudeAtlas

adr-synclisted

Regenerate the marker-managed "Architecture decisions" digest in the project's CLAUDE.md from the accepted Architecture Decision Records — one line per decision, so the project memory always reflects the ratified corpus. Shows the diff and asks before writing. Human-run by design; the model never invokes this on its own.
real-case/marvin-toolkit · ★ 0 · AI & Automation · score 71
Install: claude install-skill real-case/marvin-toolkit
# ADR Sync Project memory sync: distill the **accepted** decision records into a compact digest inside the host project's `CLAUDE.md`, between managed markers, so every future session starts aware of the ratified decisions without reading the whole corpus. This is a human gate of the ADR lifecycle (ADR-0027): the user runs it deliberately, sees the diff, and confirms before anything is written. The digest lives between: ``` <!-- marvin:adr-digest:start --> <!-- marvin:adr-digest:end --> ``` Only the block between the markers is ever regenerated — surrounding prose survives untouched. ## Scope rule **Accepted records only.** `proposed` records carry no authority yet; `superseded`, `deprecated`, and `rejected` ones carry it no longer. A record that was superseded since the last sync **drops out** of the digest on regeneration — that is the point of syncing. ## Workflow ### 1. Read the corpus Call the `adr` MCP tool with `{"action": "list"}` and keep the `accepted` records (the structured payload carries status, title, and path per record). None accepted → say so and stop; an empty digest is not worth a managed block. ### 2. Build the digest One line per accepted record — a link and a one-line summary: ```markdown | ADR | Decision | In one line | |-----|----------|-------------| | [0007](docs/adr/0007-marvin-working-directory.md) | Unified `.marvin/` working directory | Every generated service file lives under one hidden root, one subdir per command group. | ``` - L