Academic Paper Reviewer — 7-Agent Review System (Hermes Edition)

📄 License: CC BY-NC 4.0 · Copyright (c) 2026 Cheng-I Wu
🔗 Original: Imbad0202/academic-research-skills
🔄 Adaptation: Multi-agent review system implemented via delegate_task instead of Claude Code's internal agent system. All agent definitions, references, and quality standards preserved unchanged from original. This adaptation is distributed under the same CC BY-NC 4.0 license.

Quick Start

Review this paper for journal submission

Agent Team

Hermes Execution

Full Mode: 5-Panel Parallel Review

delegate_task(tasks=[
    {"goal": "Review manuscript methodology: design appropriateness, validity threats, replicability. Score 1-5.", "context": "Use agents/methodology_reviewer.md", "toolsets": ["file"]},
    {"goal": "Review evidence: citation quality, source credibility, evidence hierarchy alignment. Score 1-5.", "context": "Use agents/evidence_reviewer.md", "toolsets": ["file"]},
    {"goal": "Review argument: logical flow, claim-evidence alignment, counter-argument handling. Score 1-5.", "context": "Use agents/argument_reviewer.md", "toolsets": ["file"]},
    {"goal": "Review domain positioning: literature coverage, theoretical grounding, contribution significance. Score 1-5.", "context": "Use agents/domain_reviewer.md", "toolsets": ["file"]}
])

Editorial Decision

delegate_task(goal="Aggregate all 4 reviewer reports. Apply weighted scoring (Method 30%, Evidence 25%, Argument 25%, Domain 20%). Issue editorial decision: Accept/Minor Revision/Major Revision/Reject with justification.", context="Use agents/editor_in_chief.md", toolsets=["file"])

Revision Roadmap

delegate_task(goal="Convert editorial decision + reviewer reports into structured Revision Roadmap: prioritized action items, estimated effort, dependency mapping.", context="Use agents/revision_coach.md", toolsets=["file"])

6 Modes

Calibration Mode

Measures reviewer accuracy: FNR (False Negative Rate), FPR (False Positive Rate), AUC. Requires ground-truth labels on prior reviewed papers.

Critical Rules

1. ⚠️ Reviewers are paper-blind (don't see author info)
2. ⚠️ Every criticism must include specific actionable suggestion
3. ⚠️ Calibration mode requires 5+ ground-truth papers

Security & Privacy

Multi-agent design disclosure: This skill delegates review tasks across multiple subagents via delegate_task. Manuscript content and intermediate review outputs are processed by these agents. Use only with manuscripts you are comfortable having processed through the AI provider's delegated-agent workflow. Remove confidential material not needed for review.

Tool access: Subagents are granted only file tools for reading/writing review outputs. No terminal, web, or system tools are exposed.

Agent files: The agents/ directory contains academic peer-review prompt templates (role definitions, scoring rubrics, methodology guidelines). These are task instructions loaded as context in delegate_task calls — NOT system prompt overrides.