← ClaudeAtlas

review-dag-nodelisted

Review one DAGrail review or decision node using bounded evidence. Use when assigned semantic review, cold review, actual-artifact inspection, finding classification, or a closed approve/return decision.
CongBao/dagrail · ★ 0 · AI & Automation · score 68
Install: claude install-skill CongBao/dagrail
# Review one DAGrail node Review only the Role, Node, candidate, and evidence boundary in the assigned package. First verify that the DAGrail MCP tools are callable in this process. Skill discovery is not proof that a long-running harness loaded an upgraded MCP registration. If tools are absent, run `dagrail doctor install`, start a fresh session, or use `dagrail context`, `dagrail inspect`, `dagrail action apply`, and `dagrail pre-wait`; never construct a review transition manually. 1. Bind the assigned review Role and call `dag_context` with `view: reviewer`, `role_id`, and `node_id`. Never reuse a worker or controller session as the formal reviewer. 2. Inspect only the candidate, actual artifact, Decision, policy result, or evidence refs required by the review contract. Artifact indexes and digests are not substitutes for actual-artifact inspection when that inspection is assigned. 3. Separate semantic review, cold review, actual-artifact inspection, and deterministic admission. Do not repeat a completed responsibility from another Node. 4. Bind every finding to an owner, exact field or artifact, repair target, and concrete risk reduction. Stop at the declared scope; do not create an unbounded hardening loop. 5. Checkpoint the finding set before yielding. If only policy, validator, or fixture logic changed, apply `evidence.assess-reuse`; reuse execution only when DAGrail returns `reuse_execution`, then perform the new semantic judgment separately. 6