The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent workflow in SKILL.md that processes external data from repository files and reports generated by subagents. \n- [INDIRECT_PROMPT_INJECTION]: Ingestion points: External repository content and output from the Agent tool (subagent reports) are incorporated into the main orchestrator's context (SKILL.md). \n- [INDIRECT_PROMPT_INJECTION]: Boundary markers: The workflow in SKILL.md mandates specific delimiters and "frozen" requirement blocks in task descriptions to isolate core instructions from subagent output. \n- [INDIRECT_PROMPT_INJECTION]: Capability inventory: The skill in SKILL.md utilizes TaskCreate, TaskUpdate, and the Agent tool, providing the ability to manage state and execute complex sub-tasks. \n- [INDIRECT_PROMPT_INJECTION]: Sanitization: The instructions in SKILL.md require filtering reviewer findings against the original requirements to prevent unapproved instructions from influencing the workflow.

sdd-cc