The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is authorized to perform sequential execution of implementation and testing tools, as well as local Git operations based on the repository's configuration.
Evidence: The 'Persisted task-commit policy' section in SKILL.md outlines the procedure for making local commits following successful task verification.
Constraints: Instructions explicitly prohibit modifying global identity configurations, initializing new Git repositories, or pushing changes to remote servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided task documents and linked specifications, which represents an ingestion point for data that influences the agent's code generation logic.
Ingestion points: Read tasks, linked specs, and relevant sources (SKILL.md).
Boundary markers: Implementation is governed by a 'verification and commit policy' and requires explicit resolution of user-action blockers.
Capability inventory: Uses 'implement-task' for code changes, 'code-review' for verification, and Git for local version control.
Sanitization: Implements a sequential queue with dependency checks and terminal review gates to ensure each implementation cycle is verified before completion.
[SAFE]: The skill includes explicit security-conscious instructions for handling sensitive information in the local environment.
Evidence: 'references/blocker-readiness.md' instructs the agent to verify the presence of environment variables (e.g., in .env.local) without logging, displaying, or exfiltrating the actual values.

auto-implement