reviewer-audit

Address reviewer/editor comments by number with localized, traceable edits. Use when the user provides numbered reviewer concerns, editor proof queries, or revision lists and asks to "address comment

SKILL.md

GitHub

Reviewer Audit

Localized, traceable response to numbered reviewer/editor comments. The skill enforces scope discipline so a "fix comment 3" request never turns into a global restyle.

When to Use

  • User provides reviewer comments numbered (R1.W1, R2.O3, M1, A10, etc.) and asks to address them in the paper.
  • Editor proof queries (e.g., MDPI 27-item list, IEEE editorial sheet).
  • A revision document from advisor with itemized changes.
  • Any "fix items 5-12" style request.

Core Rules

  1. One comment, one localized change. Each numbered item maps to specific lines in specific files. Do not propagate edits beyond the cited locations.
  2. Never globally restyle on a numbered request. "Remove bold from §3.2 caption" is NOT permission to remove bold everywhere.
  3. Citations are sacred. Format/style fixes do not delete or replace citations.
  4. Audit trail. Every applied edit gets a LaTeX comment marker so it can be diffed and verified later.

Workflow

Step 1: Build the Item Table

Parse the user's input into a table:

ID Comment summary Target file Target section/lines Action type
R1.W1 clarify threat model sections/3-threat.tex §3.1, lines 12-18 rewrite paragraph
M3 shorten caption figs/main.tex Fig.4 caption shorten

If any row's target is unclear, ask the user before editing.

Step 2: Group by File and Stage

If the audit touches 3+ files or 6+ items, present a stage plan:

Stage 1: items M1-M5 in sections/abstract.tex + intro.tex (compile after)
Stage 2: items R1.W1-W4 in sections/3-threat.tex (compile after)
Stage 3: items R2.O1-O3 in sections/5-method.tex (compile after)

Wait for approval per stage. Do not bundle.

Step 3: Apply Edits with Audit Markers

For each edit, leave a LaTeX comment so the change is traceable:

% [REVIEWER R1.W1] clarified threat model: attacker has read access only
We assume the attacker observes ...

Marker format: % [REVIEWER <ID>] <one-line summary>

Step 4: Verify Compilation

After each stage, the latex-check.js PostToolUse hook runs compilation automatically. Confirm:

  • Page count delta vs. previous stage.
  • Zero new compilation errors.
  • BibTeX still resolves all \cite{}.

If compilation fails, stop and surface the error before proceeding to the next stage.

Step 5: Generate Response Table

After all stages applied, produce a markdown table mapping each ID to:

  • Comment summary
  • Files/lines touched
  • Brief description of the change
  • Page in updated PDF where the change is visible

This table feeds directly into the cover letter or rebuttal document (use review-response skill for full rebuttal drafting).

Anti-Patterns

  • ❌ "While I'm in §3.2, let me also clean up some other awkward sentences"
  • ❌ "The reviewer flagged one bold phrase, so I'll remove all bold in the paper"
  • ❌ Applying all 27 items in one batch without staged compile checks
  • ❌ Silent edits with no % [REVIEWER ID] marker
  • ❌ Substituting a flagged citation with a different one without asking

Cross-Skill References

  • For free-form rebuttal text drafting: use review-response skill.
  • For style enforcement after edits: use paper-edit skill.
  • For BibTeX/citation issues found during audit: use citation-verification skill.