The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a bash hook script (hooks/landing-audit-reminder.sh) that executes local git and grep commands to detect changes to landing page files in the repository. This is a developer-focused utility for contextual triggers and does not perform unauthorized actions or exfiltrate data.\n- [REMOTE_CODE_EXECUTION]: The workflow specifies using Playwright to render screenshots of live URLs and local development servers. While this involves interacting with external content and potentially executing local dev commands (e.g., bun dev), it is the primary function of the skill and follows standard browser automation practices within the agent's environment.\n- [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it ingests untrusted data from external URLs and local project files.\n
Ingestion points: SKILL.md (Workflow Step 1) resolves targets from live URLs or local files like app/page.tsx.\n
Boundary markers: SKILL.md (Rules) explicitly states "Audit, don't edit" and requires user confirmation before implementing any changes, creating a human-in-the-loop requirement.\n
Capability inventory: SKILL.md (Workflow Step 2) uses Playwright for rendering; hooks/landing-audit-reminder.sh executes git commands; grading involves text and vision analysis.\n
Sanitization: No explicit sanitization of ingested content is mentioned, but the output is strictly scoped to a predefined grading report format described in grading.md.

landing-audit