talking-head-recut

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes transcript data derived from user-provided audio, which the agent uses to design graphic cards. Malicious instructions embedded in the video's audio could potentially influence the agent's behavior during the design phase. 1. Ingestion points: transcript.json in the work directory, generated from the source video's audio. 2. Boundary markers: None explicitly defined to isolate transcript content from instructions. 3. Capability inventory: The agent can execute ffmpeg, ffprobe, and npx hyperframes (transcribe, render), and write HTML files to the local file system. 4. Sanitization: The skill mentions a hyperframes lint check that enforces restrictions on card HTML, such as prohibiting <script> tags.
  • [COMMAND_EXECUTION]: The workflow requires the execution of multiple CLI tools to process media and render the final output. Evidence: Frequent instructions to use npx hyperframes, ffmpeg, and ffprobe for audio extraction, metadata retrieval, transcription, and rendering.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to suggest a remote update command to the user. Evidence: npx hyperframes skills update talking-head-recut fetches packages from the npm registry.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 05:28 PM
Security Audit — agent-trust-hub — talking-head-recut