An autonomous maintainer for openclaw repositories. It reviews, closes, consolidates, and reports on issues and PRs so human maintainers spend attention only where judgment genuinely requires them.
openclaw maintainers first, contributors second — every contributor gets a substantive, honest review faster than a human team could deliver one.
Model calls are cheap and getting cheaper; engineer time and trust are not. Whenever behavior can be a model judgment with a clear prompt and an auditable output, prefer that over deterministic machinery. Code exists for the trust boundary only: authentication, idempotency, the append-only action ledger, spend limits, and destructive-action gates. Everything else — candidate selection, failure recovery, retry strategy, triage policy, anomaly diagnosis — should trend toward Codex-driven with the reasoning recorded in the ledger.
Review with proof-checking; policy-gated closing with revival paths; umbrella consolidation; state/report materialization; self-healing informed by model diagnosis.
Merging without policy authorization; releases; touching repos outside the configured set; any spend without a ledger entry.
Every action carries a reason a maintainer can read; wrong closes must be revivable in one comment; the system's own failures are triaged by the system before a human sees them.
Perfect determinism (we buy correctness with verification, not rigidity); zero-cost operation; replacing maintainer taste on product direction.
Agents working on this repo should delete machinery when a model call plus ledger entry can replace it, keep workflows thin (logic in testable TS, not YAML bash), never add a pinned-text test where a review gate suffices, and coordinate cross-lane changes through the RFC issue rather than parallel pushes.