The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes markdown files from the docs/ideas/ directory. If these files contain malicious instructions, they could potentially influence the agent's behavior during the 'List' or 'Validate' operations.
Ingestion points: Reads all files in the docs/ideas/ directory.
Boundary markers: None mentioned for file reading or parsing.
Capability inventory: The skill has file system write access and can execute shell commands to run prototypes.
Sanitization: No explicit sanitization or instruction to ignore embedded commands is present when displaying the idea list or performing the validation grill.
[DYNAMIC_EXECUTION]: The 'Prototype' operation involves the agent generating 'throwaway code' and executing it using the project's existing runners to answer logic questions.
Pattern: Writing and executing scripts in the prototypes/ folder or with a proto- prefix.
Mitigation: The instructions explicitly state that prototypes should have no persistence (memory only), be throwaway from day one, and skip non-essential polish like error handling, which reduces the complexity of the execution environment.

idea