The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool to rename output files using the mv command. This is a standard file management operation used to ensure audio files follow a specific naming convention ({slug}_part_NN.mp3) required for the storybook pipeline.
[EXTERNAL_DOWNLOADS]: The skill documentation references the ElevenLabs Model Context Protocol (MCP) available on GitHub. ElevenLabs is a well-known provider of text-to-speech services, and the reference points to their official developer resources.
[REMOTE_CODE_EXECUTION]: The skill suggests using uvx to install and run the ElevenLabs MCP. This involves fetching and executing a tool from a remote source. Because ElevenLabs is a well-known and established service, this is documented as a standard functional requirement rather than a malicious pattern.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests story content from an external file ({slug}_scenes.json in Stage 0).
Ingestion points: Reads scene text from {slug}_scenes.json (SKILL.md).
Boundary markers: None explicitly defined for the file read operation.
Capability inventory: Uses the ElevenLabs text_to_speech tool and the Bash tool for file operations.
Sanitization: No automated sanitization is described, but the skill implements a mandatory human-in-the-loop review gate (Stage 4) where the agent must present the full narration script for user approval before proceeding to audio generation, which effectively mitigates the risk of processing malicious instructions embedded in the story text.

story-narrator