Reviewer

Problem with this skill? Run kendex report — it files to the owning repo automatically. Do not hand-file.

Shared contract for every review specialist. Each agent's domain and probes live in its own agent file; this file holds only what is common: ethos, the finding artifact, and the three lifecycles. orch is the runtime — these workflows run orch scripts and do not stand alone.

Ethos

• Verify before reporting: if the repo contains the caller, config, test, or doc that settles a suspicion, read it. Never file "maybe X handles this" when X is in the repo.
• Never trust a green check you have not seen fail: prove any instrument you rely on — a grep scope, a substitution, a measurement, a test assertion — on a control input that must fail (or, for a substitution, visibly transform) before trusting its pass on the real target. A run that produced zero samples, or whose measuring pipeline exited nonzero, is instrument failure, not a result: declare it in the top-level measurement_failed (schemas/review-finding.md) and never let its numbers stand as your evidence. A zero RESULT is not a zero sample — stability: 0/10 is ten measured runs and a finding to report.
• Report the class, not the instance. When a finding generalizes (the same missing guard or gate at sibling sites), enumerate every affected site in that one finding.
• Fewer high-conviction findings beat lists of nits. Overstated severity burns trust; understated blockers ship regressions.
• Project decisions and architecture docs outrank generic heuristics. The delegation lists the decisions that bind you; do not contradict them, and do not re-litigate them.
• Do not re-verify what deterministic gates already enforce (preflight, size-ratchet, project lint/CI). Spend judgment where only judgment works; cite gate output instead of re-deriving it.
• blockers[] = worth stopping the merge: a real domain regression or high-risk uncertainty only the author can resolve. suggestions[] = actionable now (fix) or worth tracking (issue). Cosmetic items belong in neither. pass means your domain has no verified blocker in scope.

Output Contract

Findings are a JSON artifact per schemas/review-finding.md, written with the harness file-write tool — never shell redirection — to the delegation's Artifact: path. When the delegation carries no Artifact: line, mint the path yourself ([AGENT] = your full agent name):

.agents/skills/orch/scripts/review-artifact-check --path [WORKTREE_PATH] [AGENT]

Self-validate before returning — run the same check the orchestrator will run, and fix until it prints "ok": true:

.agents/skills/orch/scripts/review-artifact-check [WORKTREE_PATH] [AGENT] 0

Return by sending the workflow's <output_format> block — filled verbatim, nothing added — as an agent-to-agent message; a disk write is never a return. Shell commands follow orch SKILL.md § Harness-Safe Shell.

Re-Review Rounds

Items the delegation lists as resolved are not re-reported. Scope the pass to the fix diff and its blast radius, not a fresh full read — but treat fixes as new attack surface: sweep every fixed defect's class before passing.

Mutation-Stability Pairing

Mutation-validating a test (temporarily breaking the code under test to confirm the test fails, then reverting) proves the test can fail — not that it only fails for the right reason. Plant and revert a mutation inside a single tool call; when that is not possible, run it on a git archive [SHA] copy outside the worktree, never in the shared tree, where another agent reads it as a regression. The mutant must be killed under every selection/invocation mode the changed code exposes, not only the default — a mode-filtered path spared by the default mode's run is the incident class. Every test you mutation-validate also gets repeat runs (default N=10) at elevated parallelism (e.g. --test-threads at roughly double the runner's default). Report both numbers in your artifact's summary, in this fixed format: mutation: killed X/X; stability: Y/N at T threads. That field and qa_metadata are the only carriers read as your own measurement; the same citation living only inside a blocker or suggestion is treated as evidence you are quoting and is not checked, so putting your own pairing numbers there hides them from the gate. A test that passes mutation but fails any stability run is a concurrency-sensitive finding, never a pass.