The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs browser automation tools (Chrome DevTools MCP and Playwright MCP) to navigate and interact with live applications as part of the documentation audit process. Additionally, it uses Git commands (git worktree) to manage report generation in isolated environments, which is a recommended practice for maintaining workspace integrity.
[PROMPT_INJECTION]: As the skill is designed to process external documentation (from files, PRs, or user input), it has an inherent surface for indirect prompt injection. Maliciously crafted documentation could attempt to provide instructions to the agent during the 'walkthrough' phase.
Ingestion points: Untrusted documentation content provided via file paths, Git pull requests, or pasted text as specified in the 'Workflow' section of SKILL.md.
Boundary markers: While the skill provides a strict review rubric in references/lenses.md, it does not implement specific technical delimiters or 'ignore' instructions for the content being analyzed.
Capability inventory: The agent has access to browser-driving tools (Chrome/Playwright) and Git CLI operations as described in the 'Running the live UI safely' section.
Sanitization: The skill does not describe any specific sanitization or filtering of the ingested documentation text.
[CREDENTIALS_UNSAFE]: The skill includes a proactive security guardrail in the 'Running the live UI safely' section, explicitly instructing the agent to never read .env or other secret files to obtain credentials, and instead to request them directly from the user.

guidewright-review