PR-first adversarial review. Lean mode runs the highest-signal local Claude lenses; `deep` selects the full relevant matrix. Verified findings are posted inline before fixes, then replied to and resolved after a pushed fix.
PR-first refactor pass that runs /simplify once against an open draft PR, verifies and commits the surviving cleanups, pushes, and records them in the PR ledger. Runs at most once per PR for this engine.
High-fidelity PR-first review chain that opens or reuses a draft PR, posts verified findings inline before fixes, and runs /critique deep — preceded by /refactorpass on this engine's first pass only. Rounds 3+ run in convergence mode. Runs only when the…
Independent PR-first second opinion via the local Codex CLI. Codex finds issues read-only; Claude verifies them, posts confirmed findings inline before editing, then pushes, replies, and resolves. `verify` lets Codex run tests/build.
Post-push AI review orchestrator. Both modes fire Gemini Flash + Copilot at the same iteration watermark with staggered handling (Gemini first, Copilot folded in) — no in-skill /review. Lean caps at 2 iterations. `deep` arg bumps the cap to 4, early-exits…
Autonomous issue implementation loop with strict issue allowlisting, one linked worktree and draft PR per issue, bounded local Codex/Claude review rounds, inline thread traceability, and fresh-base validation. Use for a bounded GitHub issue queue without…
Publish a deliverable as a hosted Artifact page — choose page vs file vs comment, clear the synthetic-data gate, apply house tokens, publish or update in place. Use when work has an audience and a terminal dump or an unread markdown file would lose it — a…
Disciplined diagnosis loop for hard bugs and performance regressions. Use when something is broken, throwing, failing, flaky, or slow, or when the user says debug, diagnose, or "why is this happening".