The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided video content and narration transcripts to perform accessibility fixes, creating an attack surface for instructions embedded in external data.
Ingestion points: Processes external video files and transcript text during the audit phase.
Capability inventory: The skill modifies transcripts, adjusts video pacing, and edits visual overlays via platform-specific tools.
Boundary markers: The instructions do not define explicit delimiters to isolate untrusted data from the agent's core logic.
Sanitization: There are no mentions of sanitizing or validating transcript content before it is used to modify project data.
[COMMAND_EXECUTION]: The instructions suggest that the agent execute shell commands to install the vendor's MCP connector.
Evidence: The text includes the command claude mcp add --transport http Clueso https://connect.clueso.io/mcp as a setup step for the Claude Code assistant.
Context: While this involves automated tool installation, it targets the vendor's official infrastructure and requires user confirmation.

accessibility-pass