The Agent Skills Directory

[SAFE]: The skill implements a security-positive pattern by explicitly instructing the agent to prefer structured tools for file operations and search, limiting the use of terminal commands to specific, necessary scenarios like build processes and Git operations.
[SAFE]: All external tools and MCP namespaces referenced (Supabase, Vercel, Next.js, Microsoft Playwright) are from well-known services or trusted organizations.
[SAFE]: The skill includes a 'Verification Loop' section that mandates error checking and manual verification after file modifications, which is a defensive measure against accidental or malicious code corruption.
[INDIRECT_PROMPT_INJECTION]: The skill documents an interface where untrusted data could be processed (e.g., via fetch_webpage or browser snapshots) alongside tools that can modify the workspace. However, the skill's primary purpose is to provide safety guidelines for managing these capabilities rather than exploiting them.
Ingestion points: fetch_webpage (SKILL.md), mcp_com_vercel_ve_get_runtime_logs (mcp-reference.md), and mcp_microsoft_pla_browser_snapshot (mcp-reference.md).
Boundary markers: No specific prompt delimiters or 'ignore instructions' warnings are provided in the documentation for the data ingested via these tools.
Capability inventory: run_in_terminal, replace_string_in_file, and create_file (SKILL.md), and mcp_supabase_apply_migration (mcp-reference.md).
Sanitization: No explicit sanitization or filtering logic is suggested for external tool outputs before processing.

tools