Input: variant, prepared brief/context/selection, native reviewer results, delivery ledger. Output: the existing native review report plus a project-result envelope. Keep canonical native reports; use the existing review contracts to integrate their findings, without another finding schema.
Read references/stack-findings.md inside this skill before confidence filtering
or severity sorting. It defines how the specialist outputs enter the native review
contract while retaining original labels, source roles and evidence. Validate
required fields before treating a specialist delivery as complete.
Load the senior-review:review-quality-gates skill. Before consolidation account for every
selected reviewer, including stack specialists, failure, timeout or incomplete
delivery. Use one input finding set and delivery ledger for all dimensions;
preserve source role, native severity, premise and evidence provenance. Never merge
a failed worker into a successful result. A single final writer emits the report.
Load the skill-catalog:skill-catalog skill and run validate-usage against the
prepared knowledge_selection and all worker knowledge_usage records. Verify
every assigned scope/worker, the selection/body fingerprints and used references.
Failed, stale or missing selected knowledge is failed delivery coverage; any required
candidate gate stays unsatisfied. Keep unresolved inventory/classification/match
gaps in knowledge_coverage and the same report's coverage section. Loaded knowledge
does not prove that code paths ran, or establish a finding's premise. Preserve
knowledge provenance with each finding; generic knowledge patterns still need code
evidence and falsifiable premises. An independent derivation may use generic domain
knowledge, never shared project conclusions or peer findings.
Read references/<variant>.md inside this skill. Execute its consolidation, fresh
verification lenses, completeness check and report stages. For PR review, use the
same delivery/provenance/panel rules before producing the native PR description.
Two reviewers inheriting the same premise are an echo, not independent corroboration.
Deduplication preserves independent evidence and distinct failure modes.
Load the project-protocol:project-protocol skill to validate the snapshot and publish the
result envelope. A report can complete with explicitly degraded coverage; it cannot
claim a missing reviewer ran, or that static analysis exercised runtime behavior.