zackd-director
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches media assets from the MuAPI platform (api.muapi.ai) during the generation of keyframes, clips, and audio. These operations are authenticated via the user's environment-stored API key and are necessary for the tool's primary purpose.
- [COMMAND_EXECUTION]: Local video assembly is performed using ffmpeg and ffprobe in scripts/assemble.py. The implementation uses subprocess.run with argument lists to avoid shell execution and performs escaping on numeric values and whitelisting on transition effects to mitigate command injection risks.
- [DATA_EXFILTRATION]: Local image assets are uploaded to the MuAPI platform in scripts/provider.py. This data transfer is part of the intended animation workflow and is scoped to the skill's generated files for processing by the remote AI models.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted user topics and interpolates them into downstream AI prompts.
- Ingestion points: User-provided topic strings in scripts/script_generator.py and scripts/keyframes.py.
- Boundary markers: None present; content is directly interpolated into prompts without delimiters.
- Capability inventory: File system access (write), network operations (MuAPI uploads/downloads), and command execution (via FFmpeg subprocesses).
- Sanitization: No escaping or validation is applied to the user-supplied topic before interpolation.
Audit Metadata