The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use standard command-line tools such as npx, ffmpeg, and Playwright for video initialization, rendering, and recording. These tools are used for their intended purposes within the video production workflow.
[INDIRECT_PROMPT_INJECTION]: The reverse-engineering workflow (in references/edit-anatomy.md) identifies that transcriptions and metadata from external videos are untrusted data. It explicitly directs the agent to ignore any embedded instructions or commands within that data to prevent injection attacks.
Ingestion points: External video data (transcripts, captions, on-screen text) fetched via the watch-video or social-fetch tools.
Boundary markers: The skill uses explicit instructions to treat fetched content as untrusted data.
Capability inventory: The skill has the capability to write files to the local repository and execute shell commands (e.g., ffmpeg, npx).
Sanitization: It instructs the agent to analyze the patterns of the content while ignoring any embedded instructions, effectively sanitizing the input by filtering for stylistic cues only.
[DATA_EXPOSURE]: The skill promotes safe credential management by instructing users and agents to use environment variables for sensitive API keys (e.g., ElevenLabs). It explicitly warns against including keys in chat logs or committed files and provides remediation steps if a key is accidentally exposed.

video