| name | maintain |
| description | Run bounded, automation-safe second-brain maintenance across ingestion, linting, and safe repair. |
| disable-model-invocation | true |
| allowed-tools | Read, Glob, Grep, Write, Edit, Bash(git diff *), Bash(git status *), Bash(git log *) |
Maintain Second Brain
Cursor Automation entry point. Follow AGENTS.md and governance/policies.md. Mode from $ARGUMENTS: incremental (default) or full. If governance/automation-state.md has an empty last_successful_commit, treat incremental lint/curator scope as full for this run (then record the commit).
- Read
governance/automation-state.md and repository status. If write_run_active is true, check whether another write run is actually in progress; if not, force-unlock via maintain-state (log the reason) before continuing. Independently refuse when the worktree has dirty unrelated changes outside the expected automation branch/scope.
- Mark
write_run_active: true via knowledge-integrator (maintain-state) before mutations.
- Run the
/second-brain:ingest workflow for at most five new/absent ledger sources.
- Run
/second-brain:lint with explicit scope args: incremental verify by default, or full verify when mode is full or last_successful_commit is empty. For incremental, ensure lint computes a knowledge-filtered changed-page list for the curator.
- Run
/second-brain:repair only for unambiguous safe repairs listed in governance/policies.md (and confirmed safe by lint’s verified report).
- Require
knowledge-verifier PASS after every content mutation phase (ingest integration, repair, synthesize). Log/ledger-only early stops use verification N/A and do not need mutation PASS.
- Always clear
write_run_active: false via maintain-state on success, early stop, or failure; on success also update last_successful_commit, timestamps, ledgers, and operation log as applicable. Optionally run a final maintain-completion verify that checks the cleared flag.
- Produce a PR-ready summary: scope, sources, changed files, verification, unresolved review items, and risk-budget status from
AGENTS.md.
Honor the automation risk budget in AGENTS.md; stop and request review when exceeded. Clearing write_run_active is mandatory even when stopping for review.
