The Agent Skills Directory

[SAFE]: The skill follows a structured process for task generation, utilizing specialized internal tools (superpowers) rather than generating code directly in the prompt. This separation of concerns reduces the risk of logic errors.\n- [SAFE]: The workflow incorporates mandatory isolation via git worktrees and utilizes a review process where independent agents evaluate the generated tasks from multiple perspectives (Security, Architecture, Integration, etc.).\n- [SAFE]: While the skill ingests external project artifacts (markdown and Gherkin files) which could theoretically contain indirect prompt injections, it mitigates this surface by using explicit role-based checklists for review and requiring verification against defined spec artifacts.\n
Ingestion points: Reads proposal.md, design.md, and *.feature files to define tasks.\n
Boundary markers: The sub-agent prompt template uses clear headers and instruction blocks to separate user data from the review logic.\n
Capability inventory: Uses the Edit tool for file modifications and the Agent tool for sub-agent orchestration. Git operations are managed through platform-specific superpowers.\n
Sanitization: The process is governed by a status.yaml file that enforces prerequisite artifacts (Gherkin or Proposal) must be completed before planning can occur.

plan