The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external shell commands to facilitate its workflow.\n
It uses subprocess.run to call ffprobe for detecting audio track durations in scripts/generate_audio.py.\n
The instructions in SKILL.md direct the agent to use vercel deploy for publishing the generated content when the --deploy flag is used.\n
It also invokes other local Python scripts for text-to-speech and image generation tasks.\n- [DATA_EXFILTRATION]: The skill transmits slide narration text to the ElevenLabs API (api.elevenlabs.io) to generate audio files. This is a legitimate operation for a well-known service required for the skill's narrated presentation feature.\n- [CREDENTIALS_UNSAFE]: The script scripts/generate_audio.py is configured to retrieve the ELEVENLABS_API_KEY from a specific environment file at ~/claude-skills/elevenlabs-tts/.env. This follows the standard credential management pattern defined for this suite of skills and does not involve hardcoded secrets.\n- [PROMPT_INJECTION]: As the skill processes arbitrary user-provided content (topics, notes, research) to generate slide content and narration, it possesses an indirect prompt injection surface.\n
Ingestion points: User input via the /present command or specified input files (SKILL.md).\n
Boundary markers: Absent (No specific delimiters or instructions to ignore embedded commands are present in the slide generation prompt).\n
Capability inventory: File system writes, shell command execution (ffprobe, vercel), and network requests to ElevenLabs.\n
Sanitization: The audio generation script in scripts/generate_audio.py includes validation to prevent directory traversal via slide identifiers.

present