The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to execute system commands. Specifically, scripts/deterministic.py and scripts/audio_lane.py invoke ffmpeg and ffprobe for video and audio analysis. The scripts/sync.py script executes the meta-ads CLI to fetch advertisement metrics. These operations are core to the skill's function and operate on local data and known CLI tools.
[PERSISTENCE]: The onboarding guide in references/onboarding.md provides instructions and templates for establishing persistence. It offers the user an optional setup for a daily background synchronization task using macOS LaunchAgents (via .plist files) or Linux crontab. While these mechanisms allow the skill to run across sessions, they are explicitly presented to the user as a feature for performance optimization.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including video content and advertisement metrics, through an LLM (Gemini). This creates an attack surface where malicious data could attempt to influence the agent's output.
Ingestion points: Ad metrics fetched via the meta-ads CLI and video files processed by scripts/tag_gemini.py.
Boundary markers: The system prompt in scripts/tag_gemini.py enforces a strict JSON response structure with fixed enums to constrain the AI's output.
Capability inventory: The skill can execute shell commands (ffmpeg, meta-ads), write to the filesystem (briefs, SQLite databases), and perform network operations (Gemini API).
Sanitization: scripts/tag_gemini.py performs rigorous JSON schema validation and enum checking on the LLM's response before storing or processing attributes.
[DYNAMIC_EXECUTION]: In scripts/deterministic.py, the skill uses importlib.import_module to dynamically load the audio_lane module. This is used to handle the optional librosa dependency for advanced audio analysis, allowing the skill to remain functional even if the library is missing.

creative-signal