| name | review-findings-deduper |
| description | Consolidates duplicated findings from overlapping review passes (bots, lint, peers, subagents). Use before shipping when multiple validators produced redundant items, when triaging reviewer spam, merging PR commentary, prioritizing rework queues. |
| disable-model-invocation | true |
Review findings deduper
Inputs
Structured list or pasted markdown containing items with optional severity + file path.
Steps
- Normalize taxonomy → map synonyms to buckets: correctness, regression risk, UX, perf, accessibility, portability, observability, security.
- Dedup key → use
(file,line_range,rule_signature) triples whenever possible else (file,code_symbol,rule_signature).
- Collapse overlapping messages into one actionable paragraph with citations referencing source reviewers.
- Prioritize matrix
| Quadrant | Action |
|---|
| High severity ∩ easy fix | Immediate |
| High severity ∩ hard fix | Plan + spikes |
| Low severity ∪ nit | Backlog toggle |
| Flaky/low confidence | Mark needs-more-data |
- Emit Markdown table:
| Priority | Canonical finding | Sources merged | Owners | Acceptance check |
|-----------|---------------------|----------------|-------|---------------------|
- Optionally append distilled row to
%USERPROFILE%\Documents\Wiki\patterns\review-dedupe-log.md.
Guards
Never drop unique security regressions purely because wording differs — escalate those manually.
