The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository files such as DESIGN.md, existing plans, and project source code, which could potentially contain instructions designed to influence the agent's behavior.
Ingestion points: Reads DESIGN.md, PLAN.md, MAP.md, and codebase source files to establish context and resolve decisions.
Boundary markers: Utilizes structured markdown templates to organize data but does not implement strict isolation or escaping for embedded repository content.
Capability inventory: Capable of writing documentation files to the local repository and delegating implementation or research tasks to other specialized skills.
Sanitization: Explicitly requires human-in-the-loop (HITL) authorization for all material scope or decision changes and mandates privacy checks for evidence.
[DYNAMIC_EXECUTION]: The skill permits the creation and execution of isolated logic prototypes to verify assumptions during the planning phase.
Evidence: Instructions in references/wayfinding.md direct the agent to build runnable prototypes using 'throwaway code' to resolve behavioral uncertainties.
Context: Prototypes are intended for isolated investigation and their results are subject to human review within the conversation, which limits their use to authorized technical verification.

he-plan