The Agent Skills Directory

[SAFE]: The skill provides procedural guidance for task management and project alignment without introducing security risks.
[PROMPT_INJECTION]: The skill defines a surface for processing untrusted user input and existing project documentation during the alignment phase.
Ingestion points: User-provided requirements and historical project documents (docs/decisions/, MEMORY.md).
Boundary markers: The skill incorporates a mandatory "Requirement Alignment" step where the agent must confirm its understanding with the user via specific questions, serving as a functional boundary.
Capability inventory: The agent performs local file searches using grep and creates planning/prototype documents within the project's designated documentation directories.
Sanitization: No programmatic sanitization is defined, but the structured workflow and required user confirmation mitigate the risk of accidental instruction obedience from data.
[COMMAND_EXECUTION]: Uses the standard grep utility to search local project documentation, which is a common and safe operation in this context.

task-start