it-reelsmaker-online

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_SAFE]: The skill manages numerous sensitive API keys (OpenAI, Anthropic, Fal, etc.) with a high degree of security.
  • Evidence: scripts/reels_online.py stores keys in ~/.config/it-reelsmaker/keys.env using restrictive file permissions (mode 0600).
  • Evidence: A global scrub() function is implemented and used to redact all secret values from error messages and output text before they are displayed or logged.
  • Evidence: The SafeRedirect class in scripts/reels_online.py ensures that sensitive authorization headers are stripped if a network request is redirected to a different origin.
  • [COMMAND_EXECUTION]: The skill executes system commands for media processing and remote connectivity.
  • Evidence: scripts/matte_server.py uses subprocess to call ssh, scp, and tar for remote figure cut-outs.
  • Evidence: scripts/transcribe_online.py uses ffmpeg for audio re-encoding.
  • Evidence: All shell arguments are sanitized using shlex.quote(), and the remote SSH host is validated against a strict regex to prevent command injection.
  • [EXTERNAL_DOWNLOADS]: The skill fetches footage and assets from established providers.
  • Evidence: Downloads video clips from Pexels, Pixabay, and Openverse using HTTPS.
  • Evidence: All paid operations (generation and transcription) and large downloads require explicit user approval (the --yes flag) and check for a global REELS_OFFLINE flag.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external providers (stock metadata, AI transcripts, and translations), presenting a potential attack surface.
  • Ingestion points: Video transcripts from OpenAI/Groq, translation results from Anthropic/Gemini, and search metadata from stock/meme sites.
  • Boundary markers: The skill does not use explicit boundary delimiters, but relies on structured prompt templates to process external strings.
  • Capability inventory: High-privilege capabilities include SSH command execution, file system modification, and broad network access.
  • Sanitization: All output is passed through a scrubbing filter to prevent credential leakage; shell commands are quoted to prevent escape.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 02:38 AM