The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess module across several scripts, including scripts/optimize_skill.py, scripts/llm_cli.py, and scripts/trigger_test.py, to invoke local LLM CLI tools, manage version control via git, and list plugins. These operations are core to its management functionality.
[EXTERNAL_DOWNLOADS]: As described in SKILL.md, the skill searches for and installs new capabilities from the skills.sh registry and arbitrary GitHub repositories using the npx skills add command.
[DYNAMIC_EXECUTION]: The optimization workflow in scripts/optimize_skill.py implements a feedback loop that modifies the instruction text in SKILL.md files. It applies edits generated by LLMs to improve agent performance on specific task sets based on performance metrics.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface as it ingests and analyzes external data such as task prompts, skill documentation, and agent trajectories from third-party sources during optimization and review. It mitigates this by using structural delimiters when constructing prompts for evaluation.

skills-management