MarkMyAss

The crew comes aboard your AI workflow.

Install MarkMyAss in Your AI Workflow#

Inspect, clean and verify supported signals directly from your AI tools — free, open source, 100% local#

MarkMyAss ships as an Agent Skill: a small instruction file that teaches an AI agent to run the local MarkMyAss engine on the files and text you're working with. Once installed, your AI workflow can automatically inspect, clean and verify supported AI watermark, metadata and provenance signals — hidden Unicode, EXIF/XMP/IPTC, PDF and PNG metadata, and supported C2PA container structures — without you visiting this site.

The workflow the Skill follows is the same honest pipeline as the cleaner on this site:

The Skill's standing orders

AI creates content or a file → MarkMyAss inspects supported signals → cleans what's supported → verifies the result (with ExifTool cross-check when available) → returns the cleaned file plus a real status report.

Everything runs on your machine. Nothing is uploaded, and MarkMyAss stays free.

Technical honesty, up front: the Skill uses the exact same capability model as the rest of MarkMyAss. It removes and verifies supported signals. It does not defeat statistical model-level watermarks (Claude / Gemini / GPT — their status is always reported as UNKNOWN, because no public verifier exists), and it never claims "100% AI-undetectable." See the AI Watermark Lab for the full scored capability matrix.

A. Install from your AI product#

Different AI products use different mechanisms — these are the ones that genuinely support it today.

Claude Code#

Claude Code discovers Agent Skills from your personal skills directory (~/.claude/skills/) or a project's .claude/skills/ folder.

  1. Get the Skill folder:
git clone https://github.com/bens777/MarkMyAss
  1. Install it as a personal skill (available in every project):
mkdir -p ~/.claude/skills && cp -r MarkMyAss/skill/markmyass ~/.claude/skills/

    …or into one project only: cp -r MarkMyAss/skill/markmyass YOUR_PROJECT/.claude/skills/

  1. That's it. Ask Claude Code to "inspect this file for hidden metadata" — it will install the ghostmark CLI on first use (the Skill checks before installing) and run the inspect → clean → verify pipeline.

Claude (claude.ai web & desktop)#

claude.ai supports uploading custom Skills on paid plans (Pro, Max, Team, Enterprise), with code execution enabled:

  1. Download the MarkMyAss Skill (.zip)
  2. In claude.ai, open Settings → Features → Skills and upload the zip.
  3. Enable it, then ask Claude to inspect/clean a file in any chat where code execution is available.

Note: in the claude.ai sandbox, Claude runs the Skill's tooling inside its code-execution environment rather than on your machine.

Claude API / Agent SDK#

Building your own agent? The same Skill folder works with the Skills API (/v1/skills) and with filesystem-based skills in the Claude Agent SDK — point it at skill/markmyass/ from the repository.

ChatGPT / OpenAI#

Not currently supported as an installable skill. OpenAI's products don't use the Agent Skills format: ChatGPT's mechanism for external tools is MCP connectors (Developer Mode, paid plans), which require a hosted server — MarkMyAss is deliberately local-only and doesn't currently ship one. What you can do today: install the CLI (below) and run it on ChatGPT-produced files yourself, or have a terminal-capable agent (e.g. Codex CLI) call the same ghostmark commands. If OpenAI adds local-skill support, this page will be updated.

B. Install from the terminal#

The Skill drives the same free CLI you can use directly. One command (verified against the public repository):

pip install "git+https://github.com/bens777/MarkMyAss"

Prefer an isolated install:

pipx install "git+https://github.com/bens777/MarkMyAss"

Then the full pipeline, on anything your AI produced:

ghostmark inspect document.pdf
ghostmark clean document.pdf
ghostmark verify document.cleaned.pdf

For pasted text (hidden Unicode):

ghostmark inspect-text "your text here"
ghostmark clean-text "your text here"

Every command has --json for machine-readable output, and ghostmark clean never touches your original file.

What "supported" means here#

Signal Skill/CLI support
Hidden Unicode in text Supported — detect + remove + verify
EXIF / XMP / IPTC (JPEG, PNG, WebP) Supported — detect + remove + verify
PDF DocInfo + XMP Supported — detect + remove + verify
C2PA / Content Credentials container Partial — structural detection/removal, not cryptographic validation
Claude / Gemini / GPT statistical watermarks UNKNOWN — no public verifier exists; never claimed as removed

Proof, not promises — the Skill reports what it verified, and says UNKNOWN when it can't.

Want more than one Skill?#

MarkMyAss is the specialist crew that hunts hidden marks and ghosts in your files. Moseisley is where you get the whole crew: build your own team of AI agents and assistants to help you work, research, plan and automate — free to start.

Explore Moseisley →

← Back to the MarkMyAss cleaner · Explore the AI Watermark Lab →