Edit ML/AI paper drafts for internal consistency. Use after sections exist to align claims, terminology, figures, tables, captions, limitations, and conclusion.
Scanned 9/4/2026
Install to Claude Code
npx -y skills add NeverSight/skills_feed --skill paper-draft-consistency-editor --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Paper Draft Consistency Editor?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/neversight-paper-draft-consistency-editor)More formats (shields.io, HTML) on the badges page.
---
name: paper-draft-consistency-editor
description: Edit ML/AI paper drafts for internal consistency. Use after sections exist to align claims, terminology, figures, tables, captions, limitations, and conclusion.
argument-hint: "[paper-dir] [--mode audit|edit|report] [--scope full|claims|terms|figures|results]"
allowed-tools: Read, Write, Edit, Bash, Glob
---
# Paper Draft Consistency Editor
Check and fix whether a paper draft is internally coherent. This skill is an editor pass, not a reviewer simulation and not a first-draft writer. It makes the title, abstract, introduction, methods, results, figures, tables, captions, related work, limitations, and conclusion agree with the paper's active claims and evidence.
Use this skill for:
- title/abstract/intro/conclusion consistency
- contribution bullet to experiment mapping
- claim strength drift across sections
- terminology, method name, dataset name, baseline name, metric name, and notation consistency
- figure/table/caption/main-text alignment
- result prose that no longer matches updated numbers
- unresolved `PROVISIONAL-RESULT` or `PR-###` placeholders
- checking whether the draft follows `paper/.agent/writing-contract.md`
- producing a consistency report or directly editing narrow inconsistencies
Do not use this skill to decide the paper's positioning from scratch. Use `paper-positioning-planner` for that. Do not use it to write major new prose. Use `paper-writing-memory-manager` to record stale locations, dependency conflicts, and open writing threads found during consistency checks. Use `paper-writing-assistant` after consistency issues are identified. Do not simulate hostile reviewers; use `paper-reviewer-simulator` for acceptance-risk critique.
## Skill Directory Layout
```text
<installed-skill-dir>/
├── SKILL.md
├── references/
│ ├── consistency-rules.md
│ └── edit-protocol.md
└── templates/
└── consistency-report.md
```
## Progressive Loading
- Always read `references/consistency-rules.md`.
- Read `references/edit-protocol.md` before editing paper source.
- Use `templates/consistency-report.md` for substantial reports.
- Read `paper/.agent/writing-contract.md` when present.
- Read `paper/.agent/writing-memory/`, `paper/.agent/paper-evidence-board.md`, `paper/.agent/provisional-results.md`, and root `memory/claim-board.md` / `memory/evidence-board.md` when present.
- Read figure/table maps or visual-style files when checking figures and tables.
## Core Principles
- Preserve the selected paper story. Do not silently reposition the paper.
- Distinguish inconsistency from scientific weakness. The goal is coherence, not acceptance-risk scoring.
- Prefer the current writing contract and evidence board over local inference.
- Fix low-risk naming, reference, and wording drift directly when requested.
- For claim or evidence drift, report the issue and propose the smallest safe edit.
- Do not strengthen claims to resolve inconsistencies. Weaken, narrow, or route to missing evidence instead.
- Preserve LaTeX commands, labels, citations, math, macros, comments, and formatting unless the edit directly concerns them.
## Step 1 - Locate Draft and Contract
Find:
- paper root: `paper/`, current directory, or user-provided path
- `paper/.agent/writing-contract.md` or `.agent/writing-contract.md`
- `paper/.agent/paper-evidence-board.md`
- `paper/.agent/provisional-results.md`
- source files: `main.tex`, `paper.tex`, `sections/*.tex`, appendix, figure wrappers, table files
- root memory: `memory/claim-board.md`, `memory/evidence-board.md`, `memory/action-board.md`
If no writing contract exists, still run the consistency pass from the draft, but recommend `paper-writing-contract-planner`.
## Step 2 - Choose Mode and Scope
Modes:
- `audit`: produce a report with issues and proposed fixes
- `edit`: make narrow source edits for clear inconsistencies
- `report`: save a report under `paper/.agent/draft-consistency-report.md`
Scopes:
- `full`: all checks
- `claims`: title/abstract/intro/result/conclusion claim alignment
- `terms`: names, notation, metrics, datasets, baselines
- `figures`: figures, captions, callouts, visual jobs
- `results`: tables, numbers, result prose, provisional placeholders
Default to `audit` and `full` unless the user asks for direct edits.
## Step 3 - Build the Draft Story Map
Extract a compact map:
```markdown
## Draft Story Map
- Title promise:
- Abstract claims:
- Intro thesis:
- Contribution bullets:
- Method claims:
- Experiment claims:
- Figure/table takeaways:
- Limitation scope:
- Conclusion claims:
- Active contract:
```
Mark claim IDs such as `CLM-###` when available. If IDs are missing, use local labels like `CLM-DRAFT-001`.
## Step 4 - Run Consistency Checks
Read `references/consistency-rules.md`.
Check:
- Does the title promise match the abstract and intro thesis?
- Do abstract results appear in the experiments/results section?
- Do intro contribution bullets map to experiments, figures, tables, theorem, citation, or artifact evidence?
- Are claim strengths consistent across abstract, intro, results, conclusion, and limitations?
- Do method names, dataset names, baseline names, metric names, symbols, and notation stay consistent?
- Are figure/table captions aligned with main-text interpretation?
- Are result numbers and directions consistent across tables, captions, and prose?
- Are provisional placeholders still present?
- Does related work define the same novelty boundary as intro contributions?
- Do limitations narrow exactly the right claims?
Classify each issue:
- `blocking`: could make the draft misleading or unsupported
- `major`: materially confuses the paper story
- `minor`: naming, wording, or local flow inconsistency
- `style`: polish issue that does not affect meaning
## Step 5 - Propose or Apply Edits
Read `references/edit-protocol.md`.
For each issue, choose:
- `direct-edit`: safe local wording/name/reference fix
- `rewrite-request`: needs `paper-writing-assistant`
- `contract-update`: needs `paper-writing-contract-planner`
- `evidence-update`: needs `paper-evidence-board`
- `writing-memory-update`: needs `paper-writing-memory-manager` to record dependency conflicts, stale locations, or closed threads
- `experiment-action`: needs `experiment-design-planner`
- `diagnosis`: needs `result-diagnosis`
- `leave-note`: intentional inconsistency or unresolved decision
When editing:
- edit the smallest relevant TeX or Markdown file
- keep citations, labels, refs, macros, and math intact
- avoid broad paragraph rewrites unless explicitly requested
- do not replace provisional placeholders with unverified final prose
## Step 6 - Produce Report
Use `templates/consistency-report.md` for substantial audits.
If saving and no path is specified, use:
```text
paper/.agent/draft-consistency-report.md
```
Report:
- summary of draft story
- issue table sorted by severity
- direct edits made, if any
- unresolved consistency risks
- required contract/evidence/writing follow-ups
- provisional placeholders found
## Final Sanity Check
Before finalizing:
- no claim was strengthened without evidence
- source edits are narrow and traceable
- unresolved issues are routed to the right skill
- provisional placeholders are listed
- the final response distinguishes edits made from recommendations
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!