academic-manuscript-agents

name

academic-manuscript-agents

description

Coordinate advisory-only academic manuscript drafting, revision, resubmission, reviewer-response preparation, section-agent delegation, literature positioning, theorem or proof checking, mathematical validation, and linguistic smoothing. Use when Codex needs to organize scholarly manuscript work with a master agent and specialist sub-agents for literature review, mathematical validation, and academic prose polishing while preserving files until explicit user approval.

Academic Manuscript Agents

Overview

Use this skill to coordinate academic manuscript work through a master manuscript agent and bounded specialist sub-agents. The workflow is advisory by default: agents inspect, propose, validate, and prepare approval-ready edit plans, but they do not modify existing manuscript, data, or archival files unless the user explicitly approves a specific change.

Core Workflow

1. Ground in the project before assigning agents.

   • Read project instructions, manuscript sources, reviewer comments, theory notes, bibliographies, and reproducibility notes that are directly relevant.
   • Identify original data, archival folders, raw logs, and other read-only materials before proposing any action.
   • Record the current manuscript state, target venue if known, and the user's requested scope.

2. Use the master agent as coordinator.

   • Load references/master-agent.md for manuscript-level planning, section assignment, consistency review, and integration.
   • The master agent owns global thesis, terminology, claim strength, section dependencies, and approval-ready proposals.

3. Assign specialist agents only for bounded tasks.

   • Load references/literature-review-agent.md when citations, related work, baselines, novelty positioning, or reviewer literature concerns matter.
   • Load references/mathematical-validation-agent.md when checking definitions, assumptions, propositions, proofs, algorithms, convergence, stability, or theorem-dependent claims.
   • Load references/linguistic-smoothing-agent.md when improving academic prose while preserving technical meaning.
   • Give each specialist a narrow passage, claim, section, or research question. Ask for findings, suggested text, and risks, not direct edits.

4. Reconcile before proposing changes.

   • Compare specialist outputs for conflicts in terminology, assumptions, evidence strength, citations, and manuscript structure.
   • Keep formal results, numerical demonstrations, conjectures, and applications clearly separated.
   • Prefer modest, precise claims over novelty inflation.

5. Prepare approval-ready outputs.

   • For existing files, provide affected file list, rationale, proposed diff or explicit edit list, consistency notes, and open questions.
   • Wait for explicit user approval before applying any modification.
   • Newly created notes or derived outputs should be clearly labeled and must not overwrite original data or archival sources.

Manuscript Safety Guardrails

• Inspect before proposing edits.
• Preserve original data, archival manuscripts, source datasets, experimental logs, and raw numerical outputs.
• Do not directly edit, overwrite, filter, or delete original data.
• Do not bulk-delete files or directories.
• Treat reviewer feedback as a constraint on claim strength and evidence, not as text to dismiss.
• Distinguish formal proof from numerical demonstration and domain validation.
• Avoid overstated terms such as broad "control", "validated", "bridging paradigms", or universal application claims unless the manuscript supplies the required theory and evidence.
• For mathematical papers, require assumptions, definitions, propositions or theorems, proofs, algorithms, experiments, and limitations to agree across sections.

Recommended Coordination Pattern

Use this sequence for major revisions:

1. Master agent creates a section map and risk register.
2. Mathematical validation agent checks theory and algorithm claims.
3. Literature review agent supplies citation-ready positioning and baseline recommendations.
4. Linguistic smoothing agent polishes approved passages after technical content is stable.
5. Master agent integrates all findings into one concise proposal for the user.

For small tasks, use only the relevant specialist reference and skip broad orchestration.

Output Standards

When returning a coordinated manuscript plan or edit proposal, include:

• Goal and success criteria.
• Affected sections or files.
• Specialist findings summarized by role.
• Proposed text, patch, or edit list.
• Citation or proof risks.
• Tests or checks to run, such as LaTeX compilation, bibliography checks, theorem consistency review, or reviewer-issue mapping.
• Explicit assumptions and unresolved decisions.