| name | chatto-checkup |
| description | Run a documentation-focused maintenance checkup of the Chatto codebase. Fans out to /fdr, /adr, and /chatto-architecture, then compiles a single consolidated report. Always propose-only — no changes applied without explicit user approval. At the end, points the human at other maintenance skills (/update-project-dependencies, /chatto-security-review) they may want to run themselves. |
Chatto Checkup
The standard documentation rounds. The checkup is propose-only: it audits and reports, never auto-applies fixes.
Heavier or more opinionated work — dependency upgrades, security review — is deliberately not invoked here. The end-of-report section points the human at those skills so they can decide whether to run them.
What It Runs
Run these in parallel via the Skill tool:
/fdr (no args) — audit all FDRs against the codebase. Surface discrepancies, stale design decisions, and any user-facing features that should now have an FDR.
/adr — audit ADRs for staleness. Surface cited file paths or APIs that no longer exist; flag superseded ADRs still referenced as authoritative.
/chatto-architecture — refresh docs/ARCHITECTURE.md against the current state of core services, projections, EVT event tokens, EVT/live subject patterns, current NATS resources, ConnectRPC endpoints, and realtime websocket behavior. Report drift; propose updates.
How To Run
- Invoke all three audits in parallel.
- Compile a single consolidated report at the end. Don't dump three separate report bodies on the user.
- Present the consolidated report with three sections:
- FDR drift — FDRs that no longer match the code; candidate features for new FDRs.
- ADR drift — stale references in older ADRs.
- Architecture drift — anything in
docs/ARCHITECTURE.md that's stale, including services, projections, EVT events, subject patterns, or current resource inventories.
- End with a Recommended next step — usually one action with the biggest payoff (e.g., "land the FDR-016 update first; the rest can wait").
- Then a final Other skills you may want to run section that lists, as plain pointers (no invocation):
/update-project-dependencies — bump deps within semver and run tests
/chatto-security-review — multi-agent security audit
- Wait for user direction before applying anything. The checkup never auto-applies.
Report Format
Output a single Markdown report. Brief findings, no walls of text. For every finding, include:
- What — one sentence
- Where — file path, stream name, or FDR/ADR number
- Suggested action — one line
Anti-Patterns
- Don't apply changes without asking. The checkup is a status report, not a refactor session.
- Don't repeat raw audit output verbatim. Compile and summarize. The point of the meta-skill is to compress three reports into one.
- Don't invoke
/update-project-dependencies or /chatto-security-review. They're surfaced as pointers at the end of the report; the human invokes them.
- Don't open follow-up PRs from inside the checkup. Hand the findings back; let the user decide what to act on.