The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which influences the orchestrator's decision-making flow and subsequent tool usage.
Ingestion points: The $ARGUMENTS variable is used in SKILL.md to evaluate the current situation and determine the next workflow step.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore or escape instructions potentially embedded within the $ARGUMENTS input.
Capability inventory: The skill is capable of executing shell commands for version control and file deletion, and it delegates code-writing tasks to the task-executor sub-agent.
Sanitization: No sanitization, filtering, or validation logic is defined for the input data before it is interpolated into the decision logic.
[COMMAND_EXECUTION]: The skill explicitly directs the agent to perform operations via a shell environment.
Evidence: The execution protocol for the 'Task Execution Quality Cycle' requires the agent to 'execute with Bash' for git commits. Additionally, the 'Final Cleanup' section requires the deletion of files in the docs/plans/tasks/ directory using filesystem commands.

recipe-implement