relay-reviewlisted

## Inputs - Env: optional `RELAY_SKILL_ROOT` defaults to `skills`; examples use `PR_NUM`, `BRANCH`, and `RUN_ID`. - Files: PR diff (`/tmp/pr-diff.txt`), Done Criteria anchor, structured evaluation or legacy rubric artifacts, run manifest, and optional `/tmp/review-verdict.json`. - Sibling scripts: `${RELAY_SKILL_ROOT:-skills}/relay-review/scripts/review-runner.js`. # Relay Review ## Use when - Reviewing an executor PR against frozen Done Criteria, Verification evidence, and optional Earned Rubric anchors - Running `review-runner.js` for isolated reviewer invocation, PR comments, and manifest transitions - Producing a pass, changes-requested, or escalated relay review verdict ## Do not use when - Shaping an ambiguous task before planning — use `relay-ready` - Authoring rubrics or dispatch prompts — use `relay-plan` - Delegating initial implementation or requested fixes — use `relay-dispatch` - Merging a reviewed PR — use `relay-merge` ## Context Isolation Reviews MUST run in a fresh context: no prior planning, dispatch, or conversation history. Standard path: let `review-runner.js` invoke a supported reviewer; adapter isolation and provider-specific invocation details live in `../relay-dispatch/references/agent-adapter-platform.md`. Manual fallback also needs a new session. Do not continue from dispatch, and do not review inline in the orchestrator context. ## Setup: Establish the anchor 1. Get the PR diff and Done Criteria when preparing a manual/fallback review. Ru