| name | paper-excellence |
| description | Multi-agent paper review dispatching Econometrician, Debugger, Proofreader, and Verifier in parallel. Computes weighted aggregate score per scoring-protocol.md. Use for comprehensive quality check before submission or major milestones. |
| disable-model-invocation | true |
| argument-hint | [paper .tex path OR 'all' for full project review] |
| allowed-tools | ["Read","Grep","Glob","Write","Task","Bash"] |
Paper Excellence Review
Run all review agents in parallel for a comprehensive paper quality assessment. Computes the weighted aggregate score from scoring-protocol.md.
Input: $ARGUMENTS — path to paper .tex file, or all for full project review.
Workflow
Step 1: Identify Targets
- If
$ARGUMENTS is a .tex file: review that file
- If
$ARGUMENTS is all: review output/paper/main.tex + all scripts in code/analysis/
- Also scan for
output/slides/*.tex for auxiliary scoring
Step 2: Gather Context
Before launching agents:
- Read
.claude/rules/domain-profile.md for field-specific calibration
- Read
.claude/rules/scoring-protocol.md for weight formula
- Read any existing strategy memos in
quality_reports/
Step 3: Launch Review Agents (Parallel)
Launch up to 4 agents simultaneously via Task tool:
Agent 1: Econometrician (subagent_type: econometrician)
Review [paper.tex] through all 4 phases: claim, design validity, inference, polish.
Also check scripts in code/analysis/ for code-theory alignment.
Save report to quality_reports/[file]_econometrics_review.md
Weight: Identification 25% of aggregate score.
Agent 2: Debugger (subagent_type: general-purpose, with debugger agent instructions)
Review all scripts in code/analysis/ for code quality and correctness.
Run categories 4-12 (code quality) plus categories 1-3 (strategic) if strategy memo exists.
Save report to quality_reports/[script]_code_review.md
Weight: Code 15% of aggregate score.
Agent 3: Proofreader (subagent_type: proofreader)
Review [paper.tex] for grammar, writing quality, claims-evidence alignment,
notation consistency, and compilation. 6 check categories.
Save report to quality_reports/[file]_proofread_report.md
Weight: Paper 25% of aggregate score.
Agent 4: Verifier (subagent_type: verifier)
Run standard verification (checks 1-4): LaTeX compilation, script execution,
file integrity, output freshness.
Save report to quality_reports/[file]_verification.md
Weight: Replication 5% of aggregate score.
Step 4: Launch Talk Review (Advisory, if talks exist)
If output/slides/*.tex files exist, launch:
Storyteller Review — check notation matches paper, slide count within format range.
Discussant Review — narrative flow, visual quality, content fidelity.
Talk scores are advisory, non-blocking.
Step 5: Compute Weighted Aggregate Score
Per scoring-protocol.md, compute:
Overall = Σ (weight_k × score_k) for available components
| Component | Weight | Agent |
|---|
| Literature | 10% | (skip if no lit review exists) |
| Data | 10% | (skip if no data review exists) |
| Identification | 25% | Econometrician |
| Code | 15% | Debugger |
| Paper | 25% | Proofreader |
| Polish | 10% | (from Proofreader writing quality subscore) |
| Replication | 5% | Verifier |
If components are missing, renormalize weights over available components.
Step 6: Present Results
# Paper Excellence Report: [Title]
**Date:** [YYYY-MM-DD]
**Aggregate Score:** [XX/100] — [gate status]
## Score Breakdown
| Component | Weight | Score | Issues | Agent |
|-----------|--------|-------|--------|-------|
| Identification | 25% | XX | N | Econometrician |
| Code | 15% | XX | N | Debugger |
| Paper | 25% | XX | N | Proofreader |
| Polish | 10% | XX | N | Proofreader |
| Replication | 5% | XX | PASS/FAIL | Verifier |
## Auxiliary: Talk Scores (non-blocking)
| Talk | Score | Issues |
|------|-------|--------|
## Priority Fixes (Top 5)
1. **[CRITICAL]** [Most important — from which agent]
2. **[MAJOR]** [Second priority]
...
## Full Reports
[Links to individual report files]
Step 7: Gate Enforcement
Per quality-gates.md:
- Score >= 95 + all components >= 80: "Ready for top-5 submission."
- Score >= 90: "Ready for submission. Minor polish recommended."
- Score >= 80: "Commit-ready. Address major issues before submission."
- Score < 80: "Blocked. Must fix critical/major issues."
Principles
- Paper score is blocking. Must pass quality gates.
- Talk scores are advisory. Reported but non-blocking.
- Parallel execution. All agents run simultaneously.
- Weighted aggregation. Uses scoring-protocol.md formula, not simple average.
- Don't double-count. Same issue in multiple reports counts once.
- One unified report. User sees one priority list, not separate reports.