The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash scripts (sdd-workspace, task-brief, review-package) to manage git operations, workspace creation, and task extraction. These scripts use standard tools like git and awk on the local repository to facilitate the development workflow.
[INDIRECT_PROMPT_INJECTION]: The workflow involves reading and processing external implementation plans and code diffs. While this introduces a surface for indirect prompt injection, the skill mitigates this by using isolated subagents for implementation and review, specific file-based boundaries for task context, and multi-stage review cycles.
Ingestion points: Implementation plans (PLAN_FILE), implementer reports, and git diffs are read by scripts and agents.
Boundary markers: The skill uses task-specific brief files and report files to scope subagent context, preventing the implementer from seeing the full plan or history.
Capability inventory: The skill can execute local shell scripts (git, awk), spawn subagents with specific prompts, and write files within the .superpowers/sdd/ workspace directory.
Sanitization: The bash scripts perform basic path sanitization using basename and slug validation to prevent directory traversal within the workspace root.

subagent-driven-development