The Agent Skills Directory

[SAFE]: The skill provides a structured development methodology and lacks any high-risk operations such as arbitrary command execution, network exfiltration, or credential harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill defines a controller-subagent architecture, which introduces an inherent surface for indirect prompt injection via inter-agent communication.
Ingestion points: The controller agent is instructed to read and respond to output generated by sub-agents, including status updates (DONE, BLOCKED), technical questions, and specific review feedback from the @spec-reviewer and @code-quality-reviewer agents.
Boundary markers: The process relies on the platform-level isolation provided by the task tool; no additional boundary markers or instructions to ignore embedded commands in sub-agent output are specified.
Capability inventory: The controller agent manages task delegation, answers developer queries, and coordinates code commitments through integrated git workflows.
Sanitization: The skill does not provide specific instructions for sanitizing sub-agent output before it is processed by the main agent context.

subagent-driven-development