The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages complex workflows involving sub-agent delegation and tool usage (Read, Grep, Serena MCP). All agents are explicitly defined with readonly="true", and critical rules (DC-C001, DC-P001) strictly prohibit the modification, creation, or deletion of files, ensuring the agent remains in a non-destructive state.
[DATA_EXPOSURE]: The skill performs codebase analysis to identify existing patterns and technical constraints. These activities are internal to the requirements gathering process. While companion skills like fact-check may utilize WebSearch for external documentation verification, no mechanisms for unauthorized data exfiltration or sensitive credential harvesting were identified.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including user requests and existing codebase content, representing a potential surface for indirect injection.
Ingestion points: User request parsing (analyze phase) and codebase file reading via Read and Grep tools.
Boundary markers: The skill does not explicitly define delimiters for untrusted content, but it uses a structured AskUserQuestion tool to constrain user interactions.
Capability inventory: Sub-agent delegation, requirements document generation, and task decomposition.
Sanitization: No specific sanitization or filtering of codebase content is mentioned before it is processed by the agents.
[PROMPT_INJECTION]: The skill enforces structured interactions via the AskUserQuestion tool with 2-4 defined options, which mitigates the risk of direct prompt injection through open-ended text fields. Furthermore, rules DC-C004 and DC-C005 prioritize technical validity and evidence over user preference, providing a defense against manipulative instructions.

Define Core