paper-audit

Paper Audit Skill v3.0

Unified academic paper auditing across formats (LaTeX, Typst, PDF) and languages (English, Chinese). Runs automated checks, computes dimension scores, and optionally dispatches multi-perspective review agents.

---

Capability Summary

• Run automated paper checks across .tex, .typ, and .pdf inputs.
• Produce self-check, peer-review, gate, polish, and re-audit outputs with explicit severity and priority labels.
• Combine script findings, venue-specific checklist items, and optional agent synthesis into one report flow.
• Reuse sibling writing-skill scripts for LaTeX and Typst inputs instead of re-implementing duplicate checks.
• Audit IEEE-style pseudocode blocks for float usage, caption/label/reference hygiene, and advisory readability issues.

Triggering

Use this skill when the user wants to:

• run a pre-submission readiness audit
• simulate a reviewer-style critique
• make a pass/fail submission gate decision
• compare a revised paper against a previous audit
• audit a PDF when the source format is unavailable

Trigger it even when the user only says “check my paper”, “review this submission”, “is this ready to submit?”, or “re-audit against the old report”.

Do Not Use

• fixing the paper source as the first step when the project still fails to compile badly
• full literature research or survey drafting
• writing a paper from scratch
• template-specific LaTeX or Typst editing when the user wants direct source surgery instead of an audit report
• direct source editing, grammar fixing, or expression polishing on individual sections (use latex-paper-en, latex-thesis-zh, or typst-paper based on format and language)
• fixing compilation errors as the primary task (use the format-specific skill first)

Critical Rules

• NEVER modify \cite{}, \ref{}, \label{}, math environments, or any content listed in $SKILL_DIR/references/FORBIDDEN_TERMS.md
• NEVER fabricate bibliography entries; only verify existing .bib or .yml files
• NEVER change domain terminology without explicit user confirmation
• ALWAYS distinguish [Script] (automated) findings from [LLM] (agent judgment) assessments in output
• ALWAYS distinguish IEEE hard pseudocode violations from IEEE-safe default recommendations
• All dimension scores from scripts are indicators, not definitive judgments