The Agent Skills Directory

[COMMAND_EXECUTION]: The verification scripts mutate_gate.py and verify_rendered.py use subprocess.run to invoke the uv package manager and the playwright CLI. These commands are used to manage dependencies and install the necessary browser environment for rendering explainers.
[EXTERNAL_DOWNLOADS]: The verify_rendered.py script automatically downloads and installs the Chromium browser using Playwright on its first execution. This download originates from official distribution servers associated with the Playwright project.
[REMOTE_CODE_EXECUTION]: The skill's verification process involves executing scripts via uv run --script. This mechanism fetches and runs Python dependencies specified in PEP 723 metadata blocks within the skill's own scripts from the standard Python Package Index (PyPI).
[DYNAMIC_EXECUTION]: The skill uses the Playwright library to render generated HTML in a headless browser environment. It performs runtime JavaScript evaluation via page.evaluate() to inspect DOM properties, measure layout, and verify security invariants like Content Security Policy (CSP) enforcement.
[INDIRECT_PROMPT_INJECTION]: The skill instructions contain a 'Self-Evolving' directive, advising the AI agent to modify the SKILL.md file if it encounters errors or gate failures. This creates a potential vector where malicious input processed via the --source option could influence the agent to weaken its own security instructions or bypass verification gates.

teachback