The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill relies on user-provided task descriptions to determine which files to read and explore during the 'Recon' phase. Ingestion points: User-defined task goals in Phase 1. Boundary markers: None. Capability inventory: bash (find, grep, cat, head, python) and file:read. Sanitization: None.
[Command Execution] (HIGH): Access to bash and file reading allows for potential data exposure if the agent is manipulated into 'mapping' sensitive system paths.
[Remote Code Execution] (HIGH): The inclusion of Python in the toolset allows for arbitrary code execution, which poses a severe risk when inputs are derived from potentially malicious user prompts.

battle-plan