The Agent Skills Directory

[SAFE]: The skill performs routine development tasks such as managing git worktrees, tracking progress in local files, and ensuring git hygiene (e.g., updating .gitignore). These operations are consistent with its primary purpose of workflow orchestration.
[COMMAND_EXECUTION]: Local shell utilities (git, grep, printf) are used to orchestrate the implementation workflow. These operations are limited to the repository context and the dedicated .skills/ directory used for local artifacts.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface: (1) Ingestion points: Reads implementation instructions from tasks.md and docs/agents/project.md. (2) Boundary markers: Instructions use markdown headers and specific file paths to scope inputs, emphasizing verbatim transfer of constraints to subagents. (3) Capability inventory: The skill (via subagents) has the capability to execute tests (tdd) and write code to the local filesystem. (4) Sanitization: The skill relies on a manual user approval step for the tasks.md plan before execution begins, which acts as a safety gate for malicious instructions.

execute-plan