The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data by using WebFetch to retrieve and summarize content from third-party documentation and blog URLs referenced in the Testing Handbook. This summarized content is then interpolated into generated skill files, which could lead to indirect prompt injection if an external source contains malicious instructions.
Ingestion points: The agent-prompt.md file instructs sub-agents to read the handbook's 99-resources.md and fetch external URLs.
Boundary markers: While the skill summarizes content, it lacks explicit boundary markers or safety instructions to ignore embedded prompts in the fetched text.
Capability inventory: The skill possesses the ability to write to the file system and execute local shell commands for validation.
Sanitization: The instructions focus on stripping Hugo shortcodes and formatting, but do not specifically address adversarial input sanitization.
[EXTERNAL_DOWNLOADS]: The skill orchestrates the download of the Testing Handbook repository from GitHub (https://github.com/trailofbits/testing-handbook). This reference targets a trusted organization and is documented as a necessary step for the skill's functionality.
[COMMAND_EXECUTION]: The skill's workflow and validation procedures include the execution of local Python scripts, specifically uv run scripts/validate-skills.py, to ensure the quality and correctness of generated skills.
[DYNAMIC_EXECUTION]: The generation process involves a sub-agent architecture where prompts are dynamically constructed from templates and handbook content to drive the creation of new skill instructions.

testing-handbook-generator