video-analyser
Pass
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently executes shell commands (ffmpeg, ffprobe, tesseract, whisper) that incorporate user-provided inputs like file paths and URLs. While the script uses double quotes to mitigate simple shell injection, the execution of complex CLI tools on untrusted files remains a security surface.
- [EXTERNAL_DOWNLOADS]: Processes video files from arbitrary remote URLs and Linear ticket attachments using
curl. This is consistent with the skill's purpose but involves fetching and processing potentially malicious media files from external sources. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It extracts text and audio from external video files using OCR and transcription tools and interpolates this unverified content directly into the final analysis prompt (Step 8). An attacker could embed instructions in the video's UI or audio narration to manipulate the agent's output.
- Ingestion points: SKILL.md processes external video files via OCR (Step 6) and transcription (Step 7).
- Boundary markers: Employs XML-style tags like
<ocr_frame_N>and<audio_transcript>to delimit extracted content. - Capability inventory: Includes subprocess execution (ffmpeg, tesseract, whisper), file system writes (/tmp), and network operations (curl, Linear MCP).
- Sanitization: Lacks explicit sanitization, filtering, or instruction-stripping for the data extracted from video files before prompt construction.
Audit Metadata