The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source (video review notes posted by reviewers) which is then displayed to the agent via scripts/read-notes.mjs.
Ingestion points: scripts/read-notes.mjs fetches comments and events from the here.now API or a provided JSON file.
Boundary markers: The output of scripts/read-notes.mjs does not wrap the external content in clear delimiters or provide instructions to the agent to ignore potentially malicious prompts within the notes.
Capability inventory: The skill allows the agent to execute shell commands (ffprobe, publish.sh), write files to the filesystem (scripts/build-canvas.mjs), and make network requests.
Sanitization: While the build script sanitizes data for the HTML template, the read-back script prints raw text to the agent's console, creating a surface for prompt injection if the agent interprets the review notes as instructions.
[COMMAND_EXECUTION]: The scripts/build-canvas.mjs script executes the ffprobe utility to extract metadata (duration, fps, resolution) from video files. This is a standard and expected operation for the skill's purpose.
[EXTERNAL_DOWNLOADS]: The scripts/read-notes.mjs script performs network requests to fetch review data. By default, these requests target the here.now platform, which is the service used by this skill for storage.

video-review-canvas