# Review Consolidation

> Consolidate native review findings with evidence provenance, delivery accounting and adversarial verification. Load when consolidating code-review, team-review or pr-review results.

- Skill: `acaprino/review-consolidation` (Agent Skill, multi-file: 8 files)
- Install (CLI): `npx get-skill add acaprino/review-consolidation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/acaprino/review-consolidation/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Finance & Business
- Author: acaprino (https://skillmd.com/u/acaprino)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/acaprino/review-consolidation

---


# Review consolidation

Input: variant, prepared context/selection, native results and delivery ledger.
Output: native report and project-result envelope through existing review contracts,
without another finding schema.

Read `references/stack-findings.md` before confidence filtering or severity sorting.
Retain native labels, roles and evidence; validate required delivery fields.

Load `senior-review:review-quality-gates`. Account for every expected job and
member/scope obligation against candidate/body-bound deliveries before filtering.
Validate `member_results`, including failures, timeouts and incomplete coverage.
Share canonical finding IDs without duplicating findings; preserve source role,
native severity, premise and evidence provenance in one ledger. Failed or missing
required coverage prevents completion; never merge failure into success.
A single final writer emits the report.

Account separately for the critic and eligible verification lenses
under the host isolation/run-record mechanism; discovery cannot satisfy a check
through its job ID, file or context.

Validate knowledge usage against the prepared `knowledge_selection` and worker
`knowledge_usage` using the skill inventory actually available in the host. Verify assigned
scope/worker, selection/body fingerprints and used references. Failed/stale/missing
selected knowledge fails delivery coverage and leaves required candidate gates
unsatisfied. If no host inventory is available, record knowledge validation as a
declared gap.
Preserve inventory/classification/match gaps in `knowledge_coverage` and report
coverage. Knowledge neither proves execution nor establishes a finding's premise.
Generic patterns require source evidence and falsifiable premises. Independent
derivation may use domain knowledge, never shared project conclusions or peer findings.

Read `references/<variant>.md`; execute consolidation, fresh lenses, completeness
and reporting. PR description follows the same delivery/provenance/panel gates.
Shared premises are echoes, not independent corroboration. Deduplication preserves
independent evidence and distinct failure modes.

Use the host run-record mechanism to validate the snapshot and publish the
envelope when available; if absent, declare snapshot/publication validation as a
gap. Explicitly degraded coverage cannot claim missing reviewers ran or static
analysis exercised runtime; mandatory missing coverage keeps completion open.

## Optional human view

The final writer saves complete native Markdown once in the owned output root,
using the caller's path or variant filename. Link native details with verdict,
material findings and gaps; include full text when requested or needed for approval.
Workers deliver native reports, never separate UI or another final synthesis.

No HTML renderer exists: keep the saved native Markdown report as the
authoritative `output_path` and declare HTML/navigation/filter rendering
unavailable. Do not generate a separate UI or add analysis, findings or schema.

Preserve native sections/fields, roles/severities, evidence/confidence, premises/
provenance, verification, filtered counts, knowledge/coverage gaps and degraded or
unknown states. Link worker and verification details instead of copying them.
Economy never removes required evidence, independent premise reconstruction,
lenses or gates.

Verify rendering before delivery. Diagnose failure through provider recovery and
preserve Markdown. Persistent optional failure is explicitly reported and never
listed as successful. Requested HTML remains an open deliverable; unrequested
optional view failure leaves the native outcome intact.

