The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script using the shell to generate video content.
Evidence: python3 scripts/build_lesson.py --word love --lesson /tmp/lesson.json in SKILL.md.
The execution occurs within a local directory path (/Users/jianshuo/code/mira/search-app) and uses user-supplied strings as arguments, which could lead to command injection if the input is not strictly validated as a single word.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install an external Python package from a public registry.
Evidence: pip3 install volcengine==1.0.58.
This package is the official SDK for the Volcano Engine (BytePlus) TTS service, which is a well-known cloud provider.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where untrusted user input is processed and used to perform sensitive operations.
Ingestion points: The <word> or phrase provided by the user is ingested via various triggers.
Boundary markers: None explicitly defined in the shell command template to prevent argument injection.
Capability inventory: The skill can execute shell commands via python3 scripts/build_lesson.py.
Sanitization: The instructions advise the agent to "pick the target word" if a sentence is provided, but no technical escaping or sanitization is mandated before the string is placed into the shell command.

wjs-teaching-english