The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands, including npx, find, and grep, to search local directories for SKILL.md files. These commands are executed within the agent's environment to discover installed capabilities.
[EXTERNAL_DOWNLOADS]: The skill uses npx --yes to download and run the sshlg-skills package from the npm registry. This package is provided by the skill's author (ssheleg) to facilitate skill discovery.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to find and follow instructions from external SKILL.md files, which could lead to the execution of untrusted content if a malicious skill is present in the search paths.
Ingestion points: User task requests are translated into search concepts used to locate files (SKILL.md).
Boundary markers: None are specified for the search terms or the instructions within the discovered skills.
Capability inventory: The agent is instructed to open and follow the instructions found in matching SKILL.md files, which may include further script execution or file operations.
Sanitization: There is no explicit sanitization of the generated search concepts before they are interpolated into shell commands.
[DYNAMIC_EXECUTION]: The skill dynamically constructs shell commands by inserting agent-generated concepts into npx and grep command templates. If these concepts are influenced by malicious user input, they could potentially lead to command injection, although the instructions attempt to limit concepts to short terms.

skill-search