The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands for Git operations (e.g., git -C <worktree absolute path>) and interacts with system-level build tools including Cargo and Podman to verify implementations.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface through GitHub issues and PR comments but implements specific mitigations.
Ingestion points: The agent reads context from GitHub issue descriptions and pull request comments to guide its work.
Boundary markers: The instructions explicitly state that old issue descriptions or external comments must not override the primary implementation request or existing permissions.
Capability inventory: The skill possesses Git write access, PR management capabilities, and the ability to execute code via build tools and containers.
Sanitization: The process requires proportional testing and multi-tiered review (self, independent, or specialist) before any changes are merged, ensuring external input is verified through implementation results.
[SAFE]: The skill relies on platform-provided helper tools (such as codex-worktree, codex-delivery, and plan-change) for workspace management and delivery, which aligns with best practices for agent-led automation.

execute-plan