Academic Paper — 12-Agent Writing Pipeline (Hermes Edition)

📄 License: CC BY-NC 4.0 · Copyright (c) 2026 Cheng-I Wu
🔗 Original: Imbad0202/academic-research-skills v3.1.1
🔄 Adaptation: Multi-agent pipeline implemented via delegate_task instead of Claude Code's internal agent system. All agent definitions, references, templates, and quality standards preserved unchanged from original. This adaptation is distributed under the same CC BY-NC 4.0 license.

Quick Start

Write a paper on the impact of AI on higher education quality assurance

Agent Team

#	Agent	Phase
1	intake_agent	Phase 0: Config interview
2	literature_strategist_agent	Phase 1: Literature search
3	structure_architect_agent	Phase 2: Structure/outline
4	argument_builder_agent	Phase 3: Argument construction
5	draft_writer_agent	Phase 4: Full draft
6	citation_compliance_agent	Phase 5a: Citation check
7	abstract_bilingual_agent	Phase 5b: Bilingual abstract
8	peer_reviewer_agent	Phase 6: Peer review (max 2 loops)
9	formatter_agent	Phase 7: Output formatting
10	socratic_mentor_agent	Plan mode guide
11	visualization_agent	Figures/charts
12	revision_coach_agent	Reviewer comment parsing

Hermes Execution Pattern

All phases use delegate_task. Agent files at agents/ loaded as context.

Phase 0: Configuration

delegate_task(goal="Conduct paper configuration interview: paper type, discipline, journal, citation format, output format, language, word count.", context="Use agents/intake_agent.md")

Phase 1: Literature

delegate_task(goal="Design search strategy, screen sources, produce annotated bibliography.", context="Use agents/literature_strategist_agent.md")

Phase 2: Architecture

delegate_task(goal="Select paper structure, create detailed outline with word count allocation and evidence mapping.", context="Use agents/structure_architect_agent.md. Reference: references/paper_structure_patterns.md")

⚠️ User must approve outline before Phase 3.

Phase 3: Argumentation

delegate_task(goal="Build argument blueprint: claim-evidence chains, logical flow, counter-argument handling.", context="Use agents/argument_builder_agent.md")

Phase 4: Drafting

delegate_task(goal="Write full paper draft section by section, maintaining discipline register and word count.", context="Use agents/draft_writer_agent.md. Reference: references/writing_quality_check.md")

Phase 5: Citations + Abstract (parallel)

delegate_task(tasks=[
    {"goal": "Verify all citations: format compliance, reference list completeness, DOI checking.", "context": "Use agents/citation_compliance_agent.md", "toolsets": ["file"]},
    {"goal": "Write bilingual abstract (zh-TW + EN) with 5-7 keywords each. Independent composition, not translation.", "context": "Use agents/abstract_bilingual_agent.md. Reference: references/abstract_writing_guide.md", "toolsets": ["file"]}
])

Phase 6: Peer Review

delegate_task(goal="Simulated double-blind review on 5 dimensions: Originality(20%), Method Rigor(25%), Evidence(25%), Argument(15%), Writing(15%). Max 2 loops.", context="Use agents/peer_reviewer_agent.md")

Phase 7: Format

delegate_task(goal="Convert to target format (LaTeX/DOCX/PDF/Markdown). Apply journal formatting. Generate cover letter.", context="Use agents/formatter_agent.md. Reference: references/latex_template_reference.md")

10 Modes

Mode	Trigger	Agents
full	"Write a paper"	All 9 core
outline-only	"Paper outline"	1→2→3
revision	"Revise paper"	8→5→6
abstract-only	"Write abstract"	1→7
lit-review	"Literature review"	1→2
format-convert	"Convert to LaTeX"	9 only
citation-check	"Check citations"	6 only
plan	"Guide my paper"	1→10→3→4
revision-coach	"Parse reviews"	12 only
disclosure	"AI disclosure statement"	9 only

Critical Rules

1. ⚠️ User must confirm Paper Configuration Record before Phase 1
2. ⚠️ Max 2 revision loops; unresolved → Acknowledged Limitations
3. ⚠️ Every claim must have a citation or be from paper's own data
4. ⚠️ Zero fabricated citations — every reference verified via DOI

Handoff from deep-research

intake_agent auto-detects deep-research materials (RQ Brief, Bibliography, Synthesis) and skips redundant steps.

Security & Privacy

Multi-agent design disclosure: This skill delegates writing tasks across multiple subagents. Paper content is processed through the AI provider's delegated-agent workflow. Use only with content you are comfortable having processed by AI.

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

Agent files: The agents/ directory contains academic writing prompt templates. These are task instructions for delegate_task — NOT system prompt overrides.