The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data into persistent state files which are then used to drive a multi-agent workflow involving the Architect, Developer, and Technical Writer sub-agents.
Ingestion points: User-provided context and task specifications are captured in context.json during Step 2 of the planning workflow, as described in INTENT.md.
Boundary markers: The context.json and plan.json schemas provide structure but do not explicitly define boundary markers or instructions to treat user-provided text as data rather than instructions during sub-agent processing.
Capability inventory: The orchestration scripts invoke various sub-agents that have the capability to generate file diffs and documentation modifications. The skill uses the invoke tool to execute its primary Python logic.
Sanitization: The system uses Pydantic v2 for schema validation and includes internal reference integrity checks (Plan.validate_refs()) to ensure the plan structure is valid before implementation.
[COMMAND_EXECUTION]: The skill utilizes the invoke tool in SKILL.md to execute internal Python modules (skills.planner.orchestrator.planner and skills.planner.orchestrator.executor) that manage the complex orchestration and sub-agent dispatching logic.

planner