The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves transcribing audio from user-provided video files using ElevenLabs Scribe. The resulting text is then processed by the agent to make editing decisions (cutting, grading, animating).
Ingestion points: Transcription data enters the agent's context through takes_packed.md (generated by helpers/pack_transcripts.py).
Boundary markers: No specific boundary markers or 'ignore' instructions are used when interpolating transcript phrases into the editor sub-agent's prompt.
Capability inventory: The skill possesses capabilities to write files, make network requests to ElevenLabs, and execute shell commands via ffmpeg and ffprobe (documented in helpers/render.py, helpers/grade.py, and helpers/timeline_view.py).
Sanitization: There is no evidence of sanitization or filtering of the transcribed text before it is presented to the LLM.
[COMMAND_EXECUTION]: Several helper scripts execute shell commands to perform video processing tasks.
Evidence: helpers/render.py, helpers/grade.py, helpers/timeline_view.py, and helpers/transcribe.py use subprocess.run() to invoke ffmpeg and ffprobe for frame extraction, audio processing, and rendering. These commands are constructed using parameters derived from the project's metadata and video files.
[EXTERNAL_DOWNLOADS]: The installation and execution process involves fetching code and tools from remote sources.
Evidence: install.md instructs the user to clone the repository from GitHub and uses brew to install system dependencies. SKILL.md mentions the use of npx --yes hyperframes to download and run the HyperFrames animation engine on demand.
[PROMPT_INJECTION]: The sub-agent brief for animations includes instructions to minimize interaction.
Evidence: The prompt in SKILL.md for parallel sub-agents includes: "Do not ask questions. If anything is ambiguous, pick the most obvious interpretation and proceed." While intended to streamline parallel processing, this reduces human oversight of the sub-agent's output.

video-use