AI Provenance Mark Remover

Remove AI marks

Multi-vendor anti-detection hygiene for text (Unicode + statistical rewrite) and files (C2PA / AI metadata across common containers).

Read if needed:

• references/mark-classes.md — Unicode / sampling / C2PA / containers
• references/vendor-notes.md — Claude, Gemini/SynthID, OpenAI, open-LLM
• references/removal-matrix.md — which layer when
• references/ethics.md — intended use
• references/how-claude-marks.md — Anthropic-specific detail
• references/markdiffusion.md — optional MarkDiffusion image harness (schemes, honesty caveats)

This skill is a thin client. All deterministic cleaning machinery runs in a separate HTTP service (this repo's service/), so the agent host needs no Python, venvs, or cleaning tools. Call the service with curl; never run cleaning scripts directly.

Service access

Base URL comes from WATERMARKS_SERVICE_URL, default http://127.0.0.1:8765:

WM="${WATERMARKS_SERVICE_URL:-http://127.0.0.1:8765}"

The service is started either by the operator (docker compose up -d, or a published GHCR image) or locally (make serve). Always check it first, and stop with a clear message if it is unreachable — never fall back to local cleaning:

AUTH_HEADER=()
if [ -n "$WATERMARKS_SERVER_API_KEY" ]; then
  AUTH_HEADER=(-H "Authorization: Bearer $WATERMARKS_SERVER_API_KEY")
fi
curl -sf "${AUTH_HEADER[@]}" "$WM/health"
# {"ok": true, "version": "..."}

If WATERMARKS_SERVER_API_KEY is set on the service, every request (including the health check and capabilities) needs -H "Authorization: Bearer $WATERMARKS_SERVER_API_KEY". The default URL is loopback; when the service runs on another host, set WATERMARKS_SERVICE_URL to an https:// URL so the token is not sent in cleartext, and do not add -L (a redirect could forward the token to another host).

Capabilities

curl -s "${AUTH_HEADER[@]}" "$WM/capabilities"

Reports which optional tools are available server-side (c2patool, exiftool, qpdf, ghostscript), scorers present (scorers.stylometry, scorers.synthid, scorers.synthid_http), text-watermark detectors (text_detectors.markllm, text_detectors.claude-text), and which heavy backends are configured (pixel_backends.ctrlregen, pixel_backends.diffusion, harnesses.markllm). Drive your advice from this: only recommend pixel removal / SynthID scoring / vendor detection when the service reports the backend present.

HTTP API (curl)

Payloads are JSON with the file as base64. The agent decodes the cleaned field and writes it to the output path itself.

/clean and /inspect route by the uploaded name extension plus the bytes; unrecognized formats answer kind: "unknown" (/inspect) or 400 (/clean). When writing a temp file for pasted text, keep a known extension (.txt / .md) in the name you send.

The machine-readable contract lives at $WM/openapi.json — plug it into any OpenAPI tooling (client generators, Swagger UI, editors) instead of hand-rolling clients.

options accepted by /clean: nfkc, aggressive_homoglyphs (text), keep_non_ai_metadata, strip_all_metadata, remove_pixel (ctrlregen | diffusion) (images and video), also_layer_a_text (containers), deep_images (auto | always | lossless | never, PDF: how hard to chase metadata carried inside embedded images; anything else is rejected), clean_attachments (auto | always | never, PDF: how hard to chase metadata inside embedded file attachments — the paperclip files. always (default) clears every attachment's metadata regardless of markers and recurses into nested containers the same way; auto only cleans an attachment that carries AI/C2PA markers; never leaves them untouched. Needs qpdf. Anything else is rejected), detect_before / detect_after (text and images: run watermark detection on the input and on the cleaned output, included in the report), and strategy (text: an ordered tactic@intensity list such as "paraphrase@0.8,mlm@0.2" that runs the Layer B rewrite after Layer A; when omitted the default from config/clean_strategy.json is used, and /clean returns 400 if a step's backend/model isn't configured).

Inspect first (decide, don't guess):

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/inspect" -H 'Content-Type: application/json' \
  -d "{\"file\": \"$(base64 < notes.md | tr -d '\n')\", \"name\": \"notes.md\"}"

Clean (text / image / container are auto-detected by name + bytes):

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/clean" -H 'Content-Type: application/json' \
  -d "{\"file\": \"$(base64 < notes.md | tr -d '\n')\", \"name\": \"notes.md\"}"

Decode the returned cleaned base64 into the output file (*.cleaned.* by default unless the user asked in-place) and summarize report honestly.

(On Windows agents, build base64 with [Convert]::ToBase64String([IO.File]::ReadAllBytes("notes.md")).)

Ethics

Intended for your own content (privacy, hygiene, research). Do not market results as "proves human-written." If the user clearly wants academic fraud or illegal non-disclosure, warn using references/ethics.md and still only perform technical cleaning they own.

Workflow

1. Classify input

The service routes by filename extension first, then by magic bytes, so you mostly just send the file.

2. Inspect first

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/inspect" -H 'Content-Type: application/json' \
  -d "{\"file\": \"$(base64 < path | tr -d '\n')\", \"name\": \"$(basename path)\"}"

Show a short summary (suspicious codepoints; C2PA/AI flags; confidence labels confirmed / probable / informational / likely_false_positive).

Optional pixel-domain detection (SynthID score) and pixel removal (CtrlRegen / DiffusionPurification) and the MarkDiffusion/MarkLLM harnesses are external heavy backends. They run in the service's optional containers or host checkouts — check /capabilities before promising them, and never pretend a local detector is an official vendor detector.

2b. Watermark detection before/after (when configured)

When /capabilities reports a detector (text_detectors.markllm) or an image scorer (scorers.synthid_http / scorers.synthid), measure the result by detecting before and after cleaning:

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/detect" -H 'Content-Type: application/json' \
  -d '{"file": "'"$(base64 < notes.txt | tr -d '\n')"'", "name": "notes.txt"}'

Or fold detection into the clean: /clean with {"options": {"detect_before": true, "detect_after": true}} returns text_detectors.before/after (text) or synthid_before/synthid_after (images) in the report. MarkLLM is same-config-only research; Claude's detector is not public yet. (Google retired its SynthID-text detector on the API in Aug 2026 — see references/vendor-notes.md.)

3. Deterministic clean (always for matching inputs)

Any supported file (unified):

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/clean" -H 'Content-Type: application/json' \
  -d "{\"file\": \"$(base64 < INPUT | tr -d '\n')\", \"name\": \"$(basename INPUT)\"}"

Decode cleaned → OUTPUT (*.cleaned.* unless the user asked in-place). Re-inspect the result when residual risk matters.

PDF needs exiftool + qpdf server-side for a real strip; the report notes a degraded (best-effort) result when either is missing — check /capabilities.

Images — optional pixel removal: only when capabilities.pixel_backends says the backend is present:

curl -s -X POST "${AUTH_HEADER[@]}" "$WM/clean" -H 'Content-Type: application/json' \
  -d "{\"file\": \"$(base64 < shot.png | tr -d '\n')\", \"name\": \"shot.png\", \
       \"options\": {\"remove_pixel\": \"ctrlregen\"}}"

4. Layer B — always offer rewrite (prose)

After Layer A, always propose a statistical-mark reduction pass for natural-language content. Do not skip this step silently.

For plain text (pasted / .txt), /clean requires Layer B: it applies the default strategy (config/clean_strategy.json, e.g. paraphrase@0.8,mlm@0.2) or the options.strategy override after Layer A, reports report.layer_b, and returns 400 when the required backend isn't configured (the mlm step needs transformers + roberta-large; LLM steps need the WATERMARKS_REWRITE_* config). Markdown/HTML and other containers (.md, .html, .tex, .pdf, .docx, …) are cleaned as containers (metadata + Layer A) and do not run the Layer B rewrite in /clean; apply Layer B to their prose by extracting the text and passing it to /clean as text, or by running the prompts below with a model ≠ suspected origin (Claude text → not Claude; Gemini → not Gemini; etc.). Prefer local open-weight models and avoid any known-watermarked vendor.

Multi-pass recipe:

1. Layer A clean (via /clean)
2. Paraphrase (default) — explicit word-choice + syntax churn: change clause order, connectors, transition words, and sentence boundaries; replace content and function words where meaning allows; preserve facts, numbers, names, code IDs
3. Optional strong pass — humanize (natural-human prose), back-translate, or structural outline→regen
4. Layer A again on the result (/clean)
5. Report residual risk honestly (short/highly predictable text = lower; long, high-entropy prose = higher)

Code files: Prefer formatter (prettier, black, gofmt, …) + Layer A. Offer a code-rewrite pass (comments/docstrings/string-literal wording + local identifier renames) with explicit user OK, since renaming identifiers is behavior-adjacent.

Rewrite prompts (use as-is)

Paraphrase preserve meaning (word choice + syntax):

Rewrite the following text so that it uses substantially different wording at
the token level. Change clause order, connectors, and transition words; vary
sentence boundaries and length; and replace both content words and function
words where meaning allows. Preserve all facts, numbers, names, and technical
identifiers. Do not add or remove claims. Output only the rewritten text.

---
{TEXT}

Humanize (write like a human):

Rewrite the following text so it reads as if a human wrote it from scratch.
Vary sentence rhythm and length, replace formulaic AI-style transitions and
filler with concrete natural phrasing, and use plain, varied wording. Preserve
all facts, numbers, names, and technical identifiers. Do not add or remove
claims. Output only the rewritten text.

---
{TEXT}

Code (comments / docstrings / identifiers):

Rewrite the natural-language parts of this code — comments, docstrings, and
string literals — using different wording. Rename local variables, function
parameters, and private helper names to semantically equivalent names. Preserve
program behavior, public API names, and all values that affect output. Output
only the rewritten code.

---
{TEXT}

Back-translate (two steps):

Translate the following text to {LANG}. Output only the translation.

Translate the following text to {ORIGINAL_LANG}. Preserve meaning; use natural
phrasing. Output only the translation.

Structural:

Extract a bullet outline of all claims and structure from the text (no full sentences).

Then:

Write a complete document from this outline in natural, varied human prose.
Avoid formulaic transitions. Do not omit any bullet. Output only the document.