Skip to content
Get packs

watermarks-remover

Strips AI provenance watermarks from text and files you own, via an agent skill and a local Python service.

GitHub Download v0.7.0
Stars
25.9k
Forks
3k
Licence
MIT
Language
Python
Last update
5 days ago
Type
AI model / agent

What it is

This is an agent skill plus a small Python service that removes hidden AI provenance marks from text and files, such as invisible Unicode characters, C2PA metadata, and EXIF/XMP fields. The skill is a thin client that sends work to the service over HTTP, so the agent host does not need Python itself. The README frames it as a privacy and hygiene tool for content you own.

  • Strip invisible Unicode and bidi characters from text with the deterministic scripts
  • Clean C2PA, EXIF, or XMP metadata from PNG, JPEG, PDF, or DOCX files you own
  • Run a hook that reports or cleans marks in files an agent just wrote
  • Scan a repository with the pre-commit gate before content reaches git

Great for

  • Developers who use AI coding agents and want written files scrubbed of provenance metadata
  • Creators who want to remove embedded C2PA or EXIF data from their own images or documents before sharing
  • People who want a deterministic check on files via a PostToolUse hook or pre-commit gate
  • Teams wanting a CI check that reports actionable provenance marks as SARIF

Not ideal for

  • Removing marks from content you do not own, which the README explicitly scopes out
  • Non-technical users who cannot start a local HTTP service or edit settings files, since setup involves several steps
  • Anyone expecting statistical text watermarks to be removed deterministically, since that layer depends on model rewriting and is best-effort
Step-by-step guide: how to use watermarks-removerInstall, first run, examples, tips and FAQs
watermarks-remover alternativesKilo Code vs watermarks-removerHermes Agent vs watermarks-removerOpen SWE vs watermarks-remover

How to get it

  1. 1

    Clone and build from source

    You'll need: Python 3.10+ · An agent host such as Claude Code, Codex, or Cursor

    Install the Claude Code skill with the bundled installer

    python3 install_skill.py --skill remove-ai-marks --target claude-code

    Start the local service from the repository

    make serve
  2. 2

    Run it

    • Install the skill with the installer for your agent host
    • Start the service with make serve, or run service/scripts/server.py on Windows
    • Confirm the service is reachable at http://127.0.0.1:8765 or set WATERMARKS_SERVICE_URL
    • Ask your agent to inspect or clean a file you own using the remove-ai-marks skill
  3. 3

    Get the most out of it

    • Use the PostToolUse hook in check mode to flag marks in files the agent writes
    • Switch the hook to clean mode via WATERMARKS_HOOK_MODE=clean to strip marks automatically
    • Install the pre-commit hook to gate anything heading into git
    • Use --link when installing the skill to pick up local edits live
    • Use the clean-user-facing-text skill when you only need text cleaning without the service

Downloads

Source code (.zip)main branch · straight from GitHubRelease v0.7.0Release notes and sourceOfficial website & docsgithub.com/guillaumemeyer/watermarks-remover

Strengths

  • Deterministic cleaning for invisible characters and file metadata, not only model-based rewriting
  • Installer handles several hosts (Claude Code, Codex, Cursor) and validates skills against the Agent Skills rules
  • MIT licensed and built on the Python standard library with no dependencies

Trade-offs

  • The README states that no hook can rewrite the assistant's chat message, so chat-only text is covered only by the skill workflow
  • Statistical text watermark removal depends on the agent rewriting text, which is best-effort
  • Release v0.7.0 ships no binaries, so setup requires cloning or installing from source

How it compares

ProjectStarsTypeLevelUpdatedBest for
watermarks-remover (this)25.9kAI model / agentintermediate5 days agoDevelopers who use AI coding agents and want written files scrubbed of provenance metadata
LibreChat46.6kAppintermediate4 days agoTeams that want one shared chat interface across several AI providers
langchain149kFrameworkintermediate4 days agoPython developers building LLM-powered applications
one-api37.3kAppintermediate10 months agoDevelopers who want one OpenAI-compatible endpoint for several model providers
opencodex18.2kDev toolintermediate4 days agoDevelopers who already use Codex or Claude Code and want to try other model providers
hermes-agent260kAI model / agentintermediate4 days agoDevelopers who want a personal agent they can host on their own server or machine

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.