The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands through the --measure and --extract arguments. These commands are run locally to evaluate metrics such as performance, latency, or bundle size during the optimization loop.
[EXTERNAL_DOWNLOADS]: Examples provided in the documentation use common package managers and runtimes, including npx, bun, and uv, to fetch and execute software from well-known public registries.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the ingestion of external target files.
Ingestion points: The skill reads content from files or directories specified in the --target argument (e.g., skills, prompts, and agent definitions) to analyze their purpose.
Boundary markers: There are no documented delimiters or instructions to ignore malicious content embedded within the target files.
Capability inventory: The skill possesses high-privilege capabilities, including executing shell commands, reading arbitrary local files, and modifying files on the filesystem.
Sanitization: The skill does not implement validation for the shell commands or sanitization of the content read from target files before processing.

Optimize