The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted data from external sources.
Ingestion points: External content is retrieved via WebSearch, mcp__tavily__tavily_search, and WebFetch tools during the app research phase (Phase 2).
Boundary markers: The skill lacks explicit instructions or delimiters to isolate or ignore instructions that may be embedded in the content fetched from the web.
Capability inventory: The agent is granted powerful tools including Bash, Write, and Skill, which are used to generate and move code into the project's source directory (src/).
Sanitization: There is no defined process for sanitizing or validating the web data before it influences the generated code in the implementation phase (Phase 5).
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform filesystem operations during the finalization phase (Phase 8).
Evidence: Instructions include using cp -r to promote selected patterns to src/ and mv to archive unused patterns in _trials/_archived/.
[EXTERNAL_DOWNLOADS]: The skill uses network-enabled tools to gather design inspiration.
Evidence: Uses WebSearch and WebFetch to explore top-tier applications like Linear, Vercel, and Stripe, following the research guidelines in Phase 2.

product-inspiration