kayforkind/skill-slice/academic-pdf-redaction
Claude Code and agent skill by kayforkind/skill-slice
- community
Checking that the SKILL.md file is still in the repository.
Redact author-identifying text from academic PDFs for double-blind peer review anonymization — trigger when preparing manuscripts for blind review, anonymizing submissions, or stripping names/affiliations/emails/DOIs from PDFs. Not for scanned image-only PDFs, legal e-discovery, or general PDF merge/extract.
Install
npx vskill@latest install kayforkind/skill-slice/academic-pdf-redactionSKILL.md
The same SKILL.md was published earlier as kayforkind/skill-slice/academic-pdf-redaction.
Overview
This skill removes author-identifying information from academic papers before they enter double-blind peer review. The goal is narrow: strip the handful of strings that reveal identity (names, affiliations, contact details, venue self-references) while leaving the scientific content byte-for-byte intact so reviewers can still evaluate the work.
The skill uses PyMuPDF (fitz) to search for specific text strings and physically remove the underlying glyphs — not merely paint black boxes over them.
When to Use
- Preparing an academic paper for double-blind peer review submission.
- Anonymizing a PDF manuscript by removing author names, affiliations, emails, arXiv IDs, DOIs, and venue self-references.
- You have a text-based PDF (not a scanned image) and need targeted string-level redaction.
Prerequisites
- Python 3.10+ with PyMuPDF >= 1.24 installed. Older PyMuPDF releases (before 1.24) carry known parser CVEs that a malicious document can trigger.
- Install or verify PyMuPDF (PowerShell, Windows host):
- If the PDF is a scanned image with no extractable text layer, this skill cannot help — there are no text glyphs to search for. Those documents need pixel-level redaction after OCR (see Related skills).
- The redaction pipeline and
verify_redactionhelper are inlined in this file (Steps 2–4 and Verification). This folder does not ship helper scripts or a references pack.
Procedure
Step 1 — Identify the identifying strings to redact
Collect the literal strings and decide which regex patterns to apply. PyMuPDF's search_for() only does literal substring matching — it cannot take a regular expression. Patterns split into two groups:
Literal strings (passed straight to search_for):
- Author full names — redact the complete name ("John Smith"), never the bare surname ("Smith"). A lone surname collides with unrelated cited authors and would scrub legitimate citations.
- Affiliations — institutions and companies ("Duke University", "Acme Research").
- Venue self-references — "ICML 2024", "ICML Workshop"; naming the target venue in the body is a common deanonymisation tell.
- Acknowledgement names — people thanked by name in the Acknowledgements section.
- Equal-contribution footnotes — "Equal contribution", "* Equal contribution".
Regex patterns (matched against extracted text first, then concrete hits are looked up with search_for):
- Email addresses —
[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,} - arXiv identifiers —
arXiv:\d{4}\.\d{4,5}(?:v\d+)? - DOIs —
10\.\d{4,9}/[-._;()/:A-Za-z0-9]+
Step 2 — Run the redaction pipeline
Use redact_with_pymupdf from the implementation inlined in Step 4:
Step 3 — Three rules that keep redaction correct
- Stop at the References heading. Everything from the bibliography onward stays untouched so self-citations survive. The heading is detected as a standalone line (normalized, case-insensitive, trailing colon stripped), not a loose substring — the word "references" appears in ordinary prose, and a substring match would wrongly cut the redaction short or skip half the paper.
- Redact only specific text matches. Search for exact identifying strings and black out just those rectangles. This keeps you from removing surrounding sentences and makes the result auditable.
- Verify the output before trusting it. A bug in a single pattern can silently blank most of a page. Re-open the saved PDF and confirm the page count is unchanged and the bulk of the text (80%+) remains.
Step 4 — Reference implementation (PyMuPDF / fitz)
Full pipeline (copy into the project; this folder does not ship a helper module). The key components:
Pitfalls
Never blank the References section
Self-citations live in the bibliography, and a reviewer uses the bibliography to check how the work relates to prior art. Anonymising it corrupts the evidence the review depends on. The right way to hide self-citation signals is to neutralise giveaway phrasing in the body ("in our prior work [12]"), not to delete the reference list.
Never rely on a deprecated PDF library
As of 2026, PyMuPDF releases before 1.24 carry known parser CVEs that a malicious document can trigger. Pin a current stable release:
Never redact regions or pages — only specific strings
Region- or page-level redaction destroys the content a reviewer needs and almost always removes far more than the identifying strings.
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/academic-pdf-redaction/SKILL.md
Popularity
- 7-day trend score
- 5
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/kayforkind/skill-slice/academic-pdf-redaction)