The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub Pull Requests (title, body, comments, and diffs) which is used to generate video storyboards and scripts. This creates a surface for indirect prompt injection where a malicious PR author could attempt to influence the agent's behavior.
Ingestion points: scripts/fetch-pr.mjs retrieves PR data via the GitHub CLI; scripts/ingest.mjs processes this into a visible-text.txt brief.
Boundary markers: The workflow lacks explicit boundary markers or "ignore embedded instructions" warnings when the agent interpolates the ingested text into its planning context.
Capability inventory: The skill possesses significant capabilities, including executing shell commands (gh, ffmpeg, npx), writing to the local filesystem, and dispatching sub-agents to build HTML compositions.
Sanitization: scripts/ingest.mjs provides basic mitigation by stripping HTML comments from the PR body and enforcing character/line limits on the ingested content.
[COMMAND_EXECUTION]: The skill utilizes several CLI tools and local Node.js scripts to automate the video creation process. These include the GitHub CLI for data fetching, ffmpeg/ffprobe for audio processing, and the hyperframes CLI for project management and rendering.
Evidence: scripts/fetch-pr.mjs uses execFileSync to run gh; scripts/assemble-index.mjs uses spawnSync for ffmpeg and ffprobe; scripts/audio.mjs spawns a sibling Node.js script to handle audio generation.

hyperframes-pr-to-video