The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill methodology involves processing untrusted content from external URLs and local source code to generate reports and remediation plans.\n
Ingestion points: The agent reads data from solution_url and files within repo_path (specified in SKILL.md and templates/profile.yaml).\n
Boundary markers: The instructions include a 'Hydration-Probe' to check environment state but do not implement specific delimiters or warnings to ignore instructions embedded in audited pages.\n
Capability inventory: The skill allows the agent to write files to an audit/ folder, execute shell diagnostics (grep, find), and use browser tools to interact with web content.\n
Sanitization: The skill lacks explicit logic to sanitize or escape data extracted from audited solutions before it is interpolated into audit reports.\n- [COMMAND_EXECUTION]: The skill methodology utilizes standard shell commands like grep and find to analyze code structures for accessibility issues and performance patterns (documented in checks/A11Y-001.md and checks/PERF-002.md).\n- [EXTERNAL_DOWNLOADS]: The skill recommends executing lighthouse and vite-bundle-visualizer via npx to automate performance and asset measurements. lighthouse is an official tool from Google, a well-known service.

ux-audit