The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task files located in docs/plans/tasks/ which serve as inputs for the task-executor subagent.
Ingestion points: File content read from docs/plans/tasks/{plan-name}-task-*.md in the Task Execution Cycle.
Boundary markers: The skill does not define clear boundaries or instructions to prevent the subagent from following malicious prompts embedded within these task files.
Capability inventory: The orchestrator can invoke multiple subagents, perform git commit operations, and delete files.
Sanitization: No sanitization of the task file content is specified before processing.
[COMMAND_EXECUTION]: The skill includes instructions to execute git commit and to delete all implementation task files in the docs/plans/tasks/ directory during the Final Cleanup phase.
[DYNAMIC_EXECUTION]: The workflow involves a multi-stage process where instructions generated by one subagent (task-decomposer) are treated as executable tasks for another subagent (task-executor), creating a dynamic instruction generation and execution chain.

recipe-build