Install the complete skill directory
The repository publishes skills/media-authenticity/SKILL.md as a distribution artifact. Copy the full media-authenticity directory to .cursor/skills/media-authenticity in a project or to the documented user-level Cursor skills directory.
Install and configure neuraldefend-mcp separately if the agent will call MCP tools. The skill does not supply a package, server process, API key, or filesystem permission.
Keep the authorization gate
Before upload, the skill tells the agent to explain that the file will be sent to Neural Defend and may contain biometric or other sensitive personal data. It requires explicit authorization for the specific file and says not to infer consent from a request or from filesystem access.
The workflow also acknowledges potential billability, including some unscorable face-policy rejections. If authorization is absent or ambiguous, the agent stops.
Interpret results within the documented bounds
The skill branches on status or scored, treats rejected as unscored, and sends high risk to human review. Medium remains inconclusive, and low remains non-definitive. For video, it preserves video and audio as separate observations.
The skill includes the transaction identifier for audit correlation while excluding local paths, local filenames, media bytes, raw payloads, headers, and API keys from the response.
Evidence boundary
Limitations
- Installing the skill does not install the MCP server, issue an API key, authorize media, or guarantee that a host agent implements every instruction.
Source record
