Evidence-backed review of source code and machine-consumed implementation artifacts, including tests, configuration, schemas, migrations, build scripts, and Git workspaces, diffs, commits, PRs, or ranges. Use only when the target is code or implementation…
Find and, when mutation is explicitly authorized and repository-aware evidence supports it, remove redundant production surface such as dead or superseded code, duplicated state, unused APIs, speculative abstractions, obsolete compatibility paths,…
Resolve uncertain product or architecture decisions before implementation. Use for explicit brainstorming or design requests, or when unresolved trade-offs, blast radius, or reversal cost could materially change the implementation; clear, local, reversible…
Review a Markdown design document, consume an accepted fix queue, or independently verify accepted fixes with layered contract extraction, architecture reasoning, adversarial falsification, deterministic evidence gates, risk-based partial re-review, and…
Resolve code placement and repository-map trust from repository evidence. Use automatically for existing-repository behavior changes with unclear ownership, cross-boundary impact, entry-point or dependency changes, unfamiliar non-local scope, or missing or…
Apply lightweight safeguards when task execution involves completion claims, destructive or external-state changes, debugging, long-running or resumable work, subagent delegation, or high-risk review. Use to preserve evidence, recoverability, and user control…
Risk-aware test-driven development. Use when the user explicitly asks for TDD, test-first, red-green-refactor, or a regression test before fixing a bug; ordinary requests that merely include tests use the normal workflow.
Route substantive tasks to the smallest set of explicitly requested or strongly matching skills. Use at task start when skill selection can materially change the process.