compare

name	compare
description	Compare harness evaluation history — shows score trends, per-tier deltas, diminishing returns detection, and next grade projection.

You are performing a harness evaluation comparison. This analyzes evaluation history to show trends and improvements.

Steps

1. Get evaluation history: Run:

   HARNESS_EVAL_ROOT="${CLAUDE_PLUGIN_ROOT}" bash "${CLAUDE_PLUGIN_ROOT}/scripts/history.sh" "$(pwd)" list
   

   This returns a JSON array of past evaluations.

2. Check minimum history: If fewer than 2 evaluations exist, inform the user: "Not enough evaluation history to compare. Run /harness-eval quick or /harness-eval standard at least twice to enable comparison."

3. Get comparison data: Run:

   HARNESS_EVAL_ROOT="${CLAUDE_PLUGIN_ROOT}" bash "${CLAUDE_PLUGIN_ROOT}/scripts/history.sh" "$(pwd)" compare
   

   This returns current vs previous delta.

4. Present bilingual comparison report (English first, then ---, then Korean):

   # Harness Evaluation Comparison
   
   ## Current vs Previous
   
   | Metric | Previous | Current | Delta |
   |--------|----------|---------|-------|
   | Score | {prev_score}/10 | {curr_score}/10 | {delta} |
   | Grade | {prev_grade} | {curr_grade} | {changed?} |
   
   ## Per-Tier Changes
   
   | Tier | Previous | Current | Delta |
   |------|----------|---------|-------|
   | Basic | X/Y | X/Y | ↑/↓/→ |
   | Functional | X/Y | X/Y | ↑/↓/→ |
   | Robust | X/Y | X/Y | ↑/↓/→ |
   | Production | X/Y | X/Y | ↑/↓/→ |
   
   ---
   
   # 하네스 평가 비교
   
   ## 현재 vs 이전
   
   | 지표 | 이전 | 현재 | 변화 |
   |------|------|------|------|
   | 점수 | {prev_score}/10 | {curr_score}/10 | {delta} |
   | 등급 | {prev_grade} | {curr_grade} | {changed?} |
   
   ## 단계별 변화
   
   | 단계 | 이전 | 현재 | 변화 |
   |------|------|------|------|
   | 기본 | X/Y | X/Y | ↑/↓/→ |
   | 기능적 | X/Y | X/Y | ↑/↓/→ |
   | 견고 | X/Y | X/Y | ↑/↓/→ |
   | 프로덕션 | X/Y | X/Y | ↑/↓/→ |
   

5. Score history chart: If 3+ evaluations exist, show an ASCII bar chart:

   ## Score History
   
   eval-04-06-001  ████████░░  7.2  B
   eval-04-06-002  █████████░  7.9  B
   eval-04-06-003  █████████░  8.5  A-
   

   Use █ for filled, ░ for empty, 10 chars total width.

6. Trend analysis:

   • Diminishing returns: If the last 3+ deltas show shrinking improvements (e.g., +0.7, +0.6, +0.5), warn: "Score improvements are shrinking — further gains will require infrastructure investments (CI/CD, integration tests, performance benchmarks)."
   • Grade projection: Based on current score and trend, estimate when the next grade threshold will be reached.
   • Stalled areas: Identify tiers that haven't improved across evaluations.

7. Recommendations: Based on the comparison, suggest the highest-impact actions to continue improving.

8. Save reports to files: Save the English and Korean comparison reports as separate files:

   mkdir -p .harness-eval/reports
   

   • English report: .harness-eval/reports/eval-{YYYY-MM-DD}-{NNN}-compare-en.md
   • Korean report: .harness-eval/reports/eval-{YYYY-MM-DD}-{NNN}-compare-ko.md

   Use the Write tool to create each file. Inform the user of the saved file paths.

Error Handling

• If history.sh fails: suggest running an evaluation first
• If only 1 evaluation exists: show current score and suggest running again after improvements
• If compare returns an error: display it clearly

Tone

Be analytical and forward-looking. Focus on trajectory and momentum, not just current state.

Language

Always produce the report in both English and Korean. English section first, then a horizontal rule (---), then the Korean section. Tables, scores, and charts are identical in both sections — only the prose text (analysis, recommendations, warnings) differs.