Review Synthesis

Review Synthesis

Expected Outcome

Receive JSON outputs from multiple review perspectives and produce a consolidated review report.

Step 1: Deduplication

Multiple reviewers may flag the same root cause from different angles.

• Same location + same root cause -> Merge into one, record all perspective IDs (e.g., "CON-003, BIZ-007")
• Same root cause + different locations -> Keep separate, link via related_to
• Different root causes + same location -> Keep separate
• When merging, adopt the highest severity
• Consolidate recommendations into a single actionable item

Step 2: Priority Classification

Priority	Criteria
P0 - Blocker	Critical severity; causes data loss, security breach, or system failure
P1 - Must Fix	Major from 2+ perspectives; major from risk or scalardb perspective
P2 - Should Fix	Major from only 1 perspective; minor common across 3+ perspectives
P3 - Consider	Minor/info severity

Step 3: Quality Gate Verdict

Determined based on thresholds in ${CLAUDE_PLUGIN_ROOT}/skills/review-registry.json:

• PASS: aggregate >= 3.5, critical: 0, major <= 3, all perspectives >= 3.0
• CONDITIONAL PASS: aggregate >= 2.5, critical <= 2 (with mitigations), major <= 8
• FAIL: Below the above thresholds

Step 4: Report Generation

JSON Output (reports/review/review-synthesis.json)

{
  "review_id": "uuid",
  "verdict": "PASS|CONDITIONAL_PASS|FAIL",
  "aggregate_score": 3.8,
  "perspective_scores": {},
  "findings_summary": {"total": 0, "after_dedup": 0, "by_priority": {}, "by_severity": {}},
  "findings": [{"id": "SYN-001", "priority": "P1", "source_ids": [], "perspectives": []}],
  "conditional_items": []
}

Markdown Output (reports/review/review-synthesis.md)

Headings: Verdict -> Score Summary -> P0 Blockers -> P1 Must Fix -> P2 Should Fix -> P3 Consider

Variable Input Handling

Operates with any combination of 2-5 perspectives. Reads the weights of enabled perspectives from ${CLAUDE_PLUGIN_ROOT}/skills/review-registry.json, normalizes them, and aggregates scores.