Conversation
Deepfake detection and media safety skill using Resemble AI — detects AI-generated audio, images, video, and text with confidence scores, traces audio source platforms, applies and reads watermarks, verifies speaker identity, and extracts media intelligence (speaker, emotion, misinformation signals). Packaged as SKILL.md + LICENSE (Apache-2.0). Generated docs updated via npm start per CONTRIBUTING.md.
🔍 Skill Validator Results
Summary
Full validator output```text Found 1 skill(s) [resemble-detect] 📊 resemble-detect: 3,549 BPE tokens [chars/4: 4,083] (standard ~), 24 sections, 1 code blocks [resemble-detect] ⚠ Skill is 3,549 BPE tokens (chars/4 estimate: 4,083) — approaching "comprehensive" range where gains diminish. ✅ All checks passed (1 skill(s)) ``` |
aaronpowell
left a comment
aaronpowell
left a comment
There was a problem hiding this comment.
It might be worthwhile including a compatibility field in the frontmatter to ensure that it's surfaced up that an API token will be required, in addition to it being part of the skill itself.
Moves detailed request/response schemas from SKILL.md into references/api-reference.md, bringing the SKILL body from 557 to 282 lines (validator hard cap is 500). Core decision-making content — capability decision tree, score interpretation, workflows, red flags — stays in the body where the agent needs it at query time. Also adds a compatibility field to frontmatter per review feedback: surfaces the RESEMBLE_API_KEY requirement and the public-HTTPS-URL constraint upfront.
|
Addressing both pieces of feedback: @aaronpowell — added a `compatibility` frontmatter field: ```yaml This surfaces the API token requirement and the public-HTTPS-URL constraint upfront, independent of the body content. Skill validator (500-line body cap) — reorganized the skill:
Ready for re-review. cc @aaronpowell |
aaronpowell
left a comment
aaronpowell
left a comment
There was a problem hiding this comment.
Need to run the build script to update the README with the new reference file.
| --- | ||
| name: resemble-detect | ||
| description: Deepfake detection and media safety — detect AI-generated audio, images, video, and text, trace synthesis sources, apply watermarks, verify speaker identity, and analyze media intelligence using Resemble AI | ||
| license: MIT |
There was a problem hiding this comment.
This states that it's MIT licenses but there is an Apache LICENSE file - if it intends to be MIT can you remove the LICENSE file, or change the license field to be Apache 2.0
2 participants
Adds a new skill, resemble-detect, to the
skills/directory.What it does
Deepfake detection and media safety using the Resemble AI platform. The skill teaches Copilot to:
Built around an "Iron Law" that prevents the agent from declaring media real or fake without a completed detection result. Includes a full decision tree mapping user intent to API capability, complete endpoint reference, score interpretation guide, workflow templates, anti-patterns, and error handling.
Requirements
Process followed
Per CONTRIBUTING.md:
staged, notmainskills/resemble-detect/withSKILL.md(Apache-2.0 LICENSE bundled)npm run skill:validate— passes (✅ All 297 skills are valid)npm start— ran build, generated docs updated (docs/README.skills.mdauto-updated by CI scripts)stagedbranchPairing
Works best alongside the Resemble MCP server (hosted SSE:
https://mcp.resemble.ai/sse) which gives the agent live access to Resemble docs and OpenAPI schemas.