The Agent Skills Directory

[SAFE]: The skill serves as a specialized environment for developing and evaluating AI agent skills. Its operations, including local command execution, subagent orchestration, and the creation of local web servers for result visualization, are all aligned with its primary purpose as a developer utility.
[COMMAND_EXECUTION]: The skill utilizes subprocess calls to interface with the platform's official CLI (claude -p) for testing skill triggering and description optimization. It also uses utility commands like lsof and kill to manage local server ports, which is standard for local development tools. These actions do not represent a security threat in the context of a developer workspace.
[DYNAMIC_EXECUTION]: The toolkit dynamically generates and reads various JSON metadata files and invokes internal Python scripts to aggregate data and generate reports. This is part of the intended workflow for iterative skill improvement.
[INDIRECT_PROMPT_INJECTION]: As a testing laboratory, the skill naturally processes untrusted input in the form of test prompts and skill drafts. While this presents an attack surface for indirect prompt injection, the skill is designed for controlled evaluation, and the platform's standard guardrails for subagents apply.
[EXTERNAL_DOWNLOADS]: The web viewer loads a well-known library (SheetJS) from a reputable CDN (cdn.sheetjs.com) to enable spreadsheet visualization. This is a common and safe practice for providing specialized UI functionality.

skill-lab