feat: Structured result aggregation for multi-agent reviews

[evansenter]

Summary

Formalize the result aggregation step when multiple review agents return findings, producing a deduplicated, severity-ranked summary.

Motivation

When running /pr-review-toolkit:review-pr, 3-5 agents return independently. Currently results are manually aggregated in the conversation. A structured aggregation pattern would:

• Deduplicate overlapping findings across agents
• Rank by severity (critical > important > suggestion)
• Provide a single actionable summary with file:line references
• Track which agent found what for attribution

Inspired by the multi-agent-orchestration community skill's result aggregation patterns.

Proposed Implementation

Add an aggregation step to pr-review-toolkit:review-pr that:

1. Collects all agent results
2. Parses findings into a common schema (severity, file, line, description, agent)
3. Deduplicates by file+line proximity
4. Outputs structured summary (Critical/Important/Suggestions/Strengths)

Could be a new agent (pr-review-toolkit:aggregator) or logic in the review-pr command.

Effort

Low — pattern is well understood, just needs formalization.

---

🤖 Generated with Claude Code