The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted external data during its audit and optimization workflows.
Ingestion points: The skill ingests data from external websites and repositories during site audits and video performance evaluations (identified in SKILL.md workflows).
Boundary markers: There are no explicit instructions or delimiters to isolate untrusted web content or ignore instructions embedded within audited data.
Capability inventory: The skill allows the agent to make code patches (file-write), execute ffmpeg commands for media processing, and interact with external APIs for video conversion (found in SKILL.md, web-video.md, and sirv-video.md).
Sanitization: No sanitization or validation logic is provided for external content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the use of external scripts and media components from well-known providers.
Evidence: The sirv-video.md reference suggests including the Sirv Media Viewer script (https://scripts.sirv.com/sirvjs/v3/sirv.js?modules=video) and fetching thumbnails from YouTube's official domain (i.ytimg.com). These originate from established technology services.
[COMMAND_EXECUTION]: The skill provides specific command-line recipes for media transformation.
Evidence: web-video.md contains multiple ffmpeg command examples for stripping audio, converting GIFs, and extracting posters. While these are standard tools for the skill's domain, they represent a capability to execute shell commands.

video-optimization