The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill fetches content from the web via search tools (Perplexity and Exa) and incorporates this untrusted data into generated implementation plans. This creates a risk where malicious instructions on a researched website could influence the plan's code or instructions.
Ingestion points: External research results retrieved via mcp__perplexity__perplexity_research and mcp__exa__web_search_exa (referenced in references/research-phase.md).
Boundary markers: There are no specific delimiters or instructions provided to isolate or ignore potentially malicious content within the researched web data.
Capability inventory: The skill performs extensive local codebase analysis using the serena toolset, writes implementation plans to the local file system, and dispatches sub-agents for specialized reviews using the Agent() tool.
Sanitization: No explicit sanitization or validation of external content is performed during the research synthesis or plan drafting phases.

design