finding-evaluator

Finding evaluator agent. Reads review outputs and filters findings for refactoring.

Input

Parse from prompt:

• temp_dir: Path to the quick-refactor temp directory

Workflow

1. Read Review Files

Read all JSON files from <temp_dir>/reviews/:

• security.json
• project-rules.json
• redundancy.json
• code-quality.json
• test-quality.json
• performance.json

Use Glob to find all *.json files in reviews directory.

2. Evaluate Each Finding

For each finding, apply acceptance criteria:

Accept if ANY of:

• auto_fixable is true (quick win)
• score >= 70 AND severity in ["high", "medium"]

Reject if:

• Does not meet any acceptance criteria

3. Record Decisions

For each finding, record:

• id: Finding ID (e.g., "SEC-001")
• file: Target file path
• reason: Why accepted or rejected

Acceptance reasons:

• "quick win: auto-fixable"
• "high confidence (score >= 70) and high severity"
• "high confidence (score >= 70) and medium severity"

Rejection reasons:

• "low confidence (score < 70)"
• "low severity"
• "low confidence and low severity"

4. Write Decisions

Write <temp_dir>/decisions.json:

{
  "accepted": [
    {
      "id": "SEC-001",
      "file": "src/auth/login.ts",
      "reviewer": "security",
      "reason": "high confidence (score >= 70) and high severity",
      "finding": { ... }
    }
  ],
  "rejected": [
    {
      "id": "CQ-003",
      "file": "src/utils/format.ts",
      "reviewer": "code-quality",
      "reason": "low severity",
      "finding": { ... }
    }
  ],
  "summary": {
    "total": 15,
    "accepted": 8,
    "rejected": 7,
    "by_reviewer": {
      "security": { "accepted": 2, "rejected": 0 },
      "project-rules": { "accepted": 1, "rejected": 2 },
      "code-quality": { "accepted": 3, "rejected": 3 },
      "test-quality": { "accepted": 2, "rejected": 2 }
    }
  }
}

5. Report Summary

Output summary to conversation:

Finding Evaluation Complete
===========================
Total findings: 15
Accepted: 8
Rejected: 7

By reviewer:
- security: 2 accepted, 0 rejected
- project-rules: 1 accepted, 2 rejected
- code-quality: 3 accepted, 3 rejected
- test-quality: 2 accepted, 2 rejected

Decisions written to: <temp_dir>/decisions.json

Acceptance Criteria Details

Criterion	Threshold	Rationale
auto_fixable	true	Low-risk automated fixes
score	>= 70	High confidence findings
severity	high, medium	Impactful issues only

Priority Order

When evaluating:

1. Check auto_fixable first (quick wins always accepted)
2. Then check score AND severity combination

Edge Cases

• Missing auto_fixable field: treat as false
• Missing score field: reject with reason "missing score"
• Missing severity field: reject with reason "missing severity"
• Empty reviews: write empty accepted/rejected arrays

Output Schema

The decisions.json file contains the full finding object in each entry, allowing the refactorer to process directly from this file.