skilld
All skills
guillaumemeyer avatar

/remove-ai-marks

@e2a699b
by Guillaume Meyer (The Opinionated Man)guillaumemeyer/watermarks-remover23k stars
2,655

Remove multi-vendor AI provenance marks: invisible Unicode (Layer A), statistical text watermarks via rewrite (Layer B, always offer), and C2PA/EXIF/XMP/container metadata on PNG/JPEG/WebP/SVG/PDF/DOCX/ODT/HTML/MD/TEX. Covers Claude, Gemini/SynthID-class, OpenAI provenance, and open-LLM sampling marks. Use when the user asks to strip watermarks, remove C2PA/Content Credentials, clean AI metadata, remove invisible Unicode, anti-detect clean AI output, or runs /remove-ai-marks (aliases: /remove-claude-marks).

Ask your Agent

Use this Skill: https://skilld.dev/gh/guillaumemeyer/watermarks-remover/remove-ai-marks

This session only. Nothing lands on disk.

referenceshow-claude-marks.md

≈412 tokens on demand. Your agent reads this file only when SKILL.md points to it.

How Claude marks AI-generated content

Primary source: Anthropic Help Center (EU AI Act Article 50(2) Code of Practice).

Policy snapshot

Topic Anthropic position
New models Marking for models launched on/after 2026-08-02
Older models Transition; “in progress”
Surfaces API, Claude, Claude Code, Cowork, Tag
Regions Worldwide
Detection Third-party detection promised; docs forthcoming

Mechanism 1 — embedded text watermarks

  • Applied at the model level into the text itself (not file metadata).
  • Imperceptible; survives copy-paste; may survive light editing.
  • Weakened by paraphrase, translation, heavy edit, mixing, short text.

Likely technical class (Anthropic has not published the algorithm): statistical token-sampling watermarks (Kirchenbauer / SynthID-style). See vendor-notes.md and mark-classes.md.

Layer A scripts only remove Unicode / homoglyph carriers. Layer B (rewrite) targets statistical marks.

Mechanism 2 — C2PA on files

  • Signed Content Credentials on supported types (examples: .png, .jpg, .svg).
  • Tamper-evident while present; stripped by re-encode, metadata scrub, or many upload pipelines.
  • Inspect with c2patool when installed; strip via clean_image.py / clean_file.py / ExifTool.

Caveats (Anthropic)

  • Detected mark ⇒ content may have been processed by Claude — not proof of sole authorship.
  • No mark ≠ human-only origin.
  • Proofreading / translate / summarize can stamp human material.

Source: SKILL.md on GitHub

Third-party checks

2 warnings6d3 checks · Risk MEDIUM
  • Gen Agent Trust Hub6d

    The skill cleans AI watermarks by sending file data to an external, user-definable service via network requests. It includes instructions to download and execute code from untrusted third-party repositories on GitHub for advanced cleaning features.

  • Socket6d

    1 alert: gptAnomaly

  • Snyk6d

    Risk: LOW · No issues

Provenance

Signed by skilld at e2a699b. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 5 days ago.

Activeupdated 3 weeks ago

README badge

README badge for guillaumemeyer/watermarks-remover/remove-ai-marks