An agent skill for deepfake detection and media intelligence — powered by Resemble AI.
Give any AI agent the ability to detect AI-generated audio, images, video, and text, analyze completed detection results, and tell people from AI agents on a website, using Resemble's Detect, Intelligence, and Agent Detection APIs.
Works with any agent that supports markdown skills:
| Agent | Install Method |
|---|---|
| Claude Code | npx skills add resemble-ai/detect-skill or copy to .claude/skills/ |
| OpenClaw (formerly Clawdbot) | Copy to skills directory or import via built-in skill loader |
| Hermes Agent | Add to skills directory — Hermes will auto-discover and self-improve on it |
| Cursor | Add to .cursor/skills/ or project rules |
| GitHub Copilot | Add to .github/copilot-instructions.md or reference in prompt |
| Windsurf | Add to project rules |
| Gemini CLI | Add to .gemini/skills/ |
Via skills.sh (recommended):
npx skills add resemble-ai/detect-skillManual: Copy SKILL.md into your agent's skills directory.
This skill teaches your agent to use Resemble's Detect and Intelligence APIs directly:
| Capability | Description |
|---|---|
| Deepfake Detection | Analyze audio, image, and video for synthetic manipulation with labels, scores, status, and visualizations |
| Direct Uploads | Submit local/private files directly to POST /detect with multipart upload, or use secure upload tokens for larger/non-public media |
| Audio Source Tracing | Identify which AI platform, such as ElevenLabs or Resemble, may have synthesized detected fake audio |
| Intelligence | Extract speaker info, emotion, transcription, misinformation signals, abnormalities, and image/video context |
| Detect Intelligence | Ask natural-language follow-up questions about completed detection results |
| Detect Agents | Run a managed multi-step investigation (document, claim, identity, evidence) that streams a verdict over SSE |
| Text Detection | Check whether writing — essays, emails, posts, reviews, comments — was generated by an AI model, with a prediction and confidence |
| Agent Detection | Find out whether a website's visitors are people or AI agents: set up a site, install its one-line snippet, and read how much traffic comes from agents |
- A Resemble AI API key
curlfor direct API calls- One of these media inputs for detection:
- a public HTTPS URL,
- a local file upload up to 150 MB, or
- a secure upload token for larger/non-public media
- For text detection: the text itself, at least 25 words (the detector does not score shorter text)
This skill is built around direct Resemble REST API calls. Agents do not need an MCP server to run detection workflows; they can call the API with curl using a Resemble API key.
export RESEMBLE_API_KEY="..."
BASE_URL="https://app.resemble.ai/api/v2"
curl --request POST "${BASE_URL}/detect" \
-H "Authorization: Bearer ${RESEMBLE_API_KEY}" \
-H "Prefer: wait" \
-H "Content-Type: application/json" \
--data '{
"url": "https://example.com/media.mp4",
"intelligence": true,
"visualize": true,
"audio_source_tracing": true,
"zero_retention_mode": true
}'For private/local media, POST /detect also supports direct multipart/form-data file upload up to 150 MB. For larger or non-public media, use the Secure Upload flow and pass the returned media_token into POST /detect. See SKILL.md for copy-pasteable workflows.
Text detection is its own endpoint. Send at least 25 words and allow a long timeout, since the first request after idle can take a few minutes while the model loads:
curl --request POST "${BASE_URL}/text_detect" \
-H "Authorization: Bearer ${RESEMBLE_API_KEY}" \
-H "Prefer: wait" \
-H "Content-Type: application/json" \
--max-time 320 \
--data "$(jq -n --arg text "$TEXT" '{text: $text}')"The result carries prediction (ai or human) and confidence (how sure the model is of that prediction). Text under 25 words is rejected rather than guessed.
If your agent supports MCP, you can still pair this skill with the Resemble MCP server for live documentation and endpoint schema lookup. MCP is optional; it is not required for the skill's Detect or Intelligence workflows.
Hosted endpoint:
https://mcp.resemble.ai/sse
See the Resemble MCP README for per-agent config snippets.
The skill is a single markdown file (SKILL.md) that provides your AI agent with:
- Decision tree — maps user intent to Detect, Intelligence, Detect Intelligence, Detect Agents, Text Detection, or Agent Detection endpoints
- Direct API examples — curl-first workflows for URL, file upload, secure upload, and polling
- Score interpretation — how to read and present media scores and text
prediction+confidencepairs - Workflow templates — full media forensics and quick authenticity checks
- Red flags — anti-patterns the agent should catch in its own reasoning
- Error handling — common status codes with cause and resolution
Agents like Hermes Agent with self-improving skill systems will automatically refine their use of this skill over time. OpenClaw's 100+ prebuilt skills ecosystem makes it a natural fit — drop detect.md in and it works alongside existing skills immediately.
Once installed, try asking your agent:
- "Is this audio file a deepfake?"
- "Analyze this video for AI manipulation and tell me what platform might have generated it."
- "Run detection with intelligence on this image URL."
- "Ask a follow-up question about this completed detection result."
- "What can you tell me about this audio — speaker, emotion, language, any abnormalities?"
- "Was this essay written by ChatGPT?"
- "Check these product reviews for AI-generated text."
- "Add Agent Detection to our Next.js site."
- "How much of our signup traffic last week was AI agents?"
- Resemble AI — Platform
- Detect API Documentation — Deepfake detection docs
- Submit Detection Job —
POST /detect - Intelligence Documentation — Media intelligence docs
mcp.resemble.ai/mcp— Optional hosted MCP endpoint for docs/schema lookup- resemble-ai/resemble-mcp — Optional MCP docs server
- skills.sh — The Open Agent Skills Ecosystem
- OpenClaw — Open-source AI agent (formerly Clawdbot)
- Hermes Agent — Self-improving AI agent by Nous Research
Apache-2.0