The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates complex workflows by executing local Python scripts via subprocess calls. These scripts handle benchmark aggregation, skill packaging, and trigger evaluation using the claude CLI. This behavior is central to the skill's purpose as a developer tool.\n- [EXTERNAL_DOWNLOADS]: The interactive evaluation viewer (eval-viewer/viewer.html) loads the SheetJS library from cdn.sheetjs.com at runtime to enable the inline rendering of Excel files. This is a standard use of a well-known service for data visualization.\n- [DYNAMIC_EXECUTION]: To evaluate triggering accuracy, the skill dynamically generates and removes command files within the project's .claude/commands/ directory. This allows the agent to simulate the presence of new skills during testing.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data such as user feedback and subagent transcripts to perform iterative improvements. While this creates an attack surface for indirect prompt injection, it is managed through instructions designed for high-capability agents and is necessary for the skill's stated functionality.

skill-creator