The Agent Skills Directory

[SAFE]: The skill is a well-documented orchestrator for development tasks. It uses isolation principles to separate task implementation from review and controller logic, which is a security best practice for agentic workflows. It explicitly instructs the agent to stop for security-sensitive actions or irreversible operations.
[COMMAND_EXECUTION]: The skill utilizes local bash scripts and git commands to manage plan artifacts and track progress. These operations are properly validated using git rev-parse --verify and are scoped to a designated git-ignored workspace directory (.superpowers/sdd/).
[INDIRECT_PROMPT_INJECTION]: The skill processes implementation plans which serve as the primary source of instructions for subagents.
Ingestion points: External plan files (PLAN_FILE) are parsed by scripts/task-brief to create task-specific requirement files.
Boundary markers: Prompts for implementers and reviewers use file paths to point to requirements, separating task content from the agent's core instructions.
Capability inventory: Subagents can perform file writes (reports), git operations (commits, diffs), and execute test suites as defined in the project.
Sanitization: Shell scripts use proper variable quoting to prevent command injection from plan-derived strings.

subagent-driven-development