The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external inputs such as code changes, shared data representations, configuration files, and persisted records. This creates a surface where instructions embedded within the analyzed data could potentially influence the agent's behavior.
Ingestion points: The skill explicitly consumes 'caller's frozen target, context, and scope', 'shared data representations', 'configuration', 'persisted records', and 'messages consumed across language or process boundaries' (SKILL.md).
Boundary markers: While the skill instructs the agent to 'Establish the Question' to scope the investigation, there are no explicit instructions for using delimiters or boundary markers to isolate untrusted data.
Capability inventory: The skill utilizes file reading, connection searching, and the execution of targeted checks or temporary scripts (SKILL.md).
Sanitization: There is no mention of sanitizing or escaping content from external sources before processing or interpolation.
[DYNAMIC_EXECUTION]: The skill instructs the agent to create and run executable code to verify its findings.
Evidence: The instructions state to 'Prefer a focused test, temporary script, or application reproduction for a consequential assumption that inspection cannot settle' and to 'Exercise the real path' (SKILL.md).
Safety Mitigation: The skill follows best practices by instructing the agent to 'Keep temporary checks and their outputs isolated from reviewed files and live systems' (SKILL.md).

blast-radius