The Agent Skills Directory

[COMMAND_EXECUTION]: The /skill setup and /skill scan subcommands instruct the agent to execute shell scripts for directory management and metadata extraction. Specifically, it uses find with -exec sh -c to traverse the filesystem and run commands like grep, sed, and stat on discovered files.
[EXTERNAL_DOWNLOADS]: The 'Import Skill' functionality allows the agent to fetch content from arbitrary user-provided URLs (e.g., GitHub Gists). This allows the introduction of external markdown files into the local skill environment without pre-defined source restrictions.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted data from external URLs and existing conversation history to create or update skill files. Malicious instructions embedded in these sources could be saved as persistent skills and influence the agent's behavior during future sessions.

skill