The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external websites and codebases, creating a surface for indirect prompt injection. This is mitigated by explicit instructions to remain in an analytical mode and avoid code modifications. \n
Ingestion points: Auditing live URLs via agent-browser and codebase files via Grep (SKILL.md, Phase 3). \n
Boundary markers: Shared instructions in references/agent-prompts.md explicitly state: "You are auditing DESIGN QUALITY. Analysis only — never edit code." \n
Capability inventory: Access to tools like Bash (git, gh, agent-browser), Read, Glob, and Task for spawning sub-agents. \n
Sanitization: All findings are required to be mapped to structured, authoritative design criteria files (references/criteria-foundations.md and references/criteria-interaction.md). \n- [COMMAND_EXECUTION]: The skill uses the Bash tool to execute git, gh, and agent-browser commands. These are utilized for their intended functions, such as fetching PR files or capturing website snapshots for the audit.

review-design