The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs standard Git operations (including git pull, git checkout, and git commit) and utilizes the GitHub CLI (gh) for issue management. These operations are integrated into the feature development workflow and are restricted to the local repository context.
[REMOTE_CODE_EXECUTION]: The skill dispatches tasks to subagents using platform-specific mechanisms like the Agent tool. This is a common architectural pattern for complex agent workflows to manage context and provide task isolation.
[PROMPT_INJECTION]: The skill processes external plans and user prompts via the <input_document> boundary. It implements a multi-phase triage and clarification process in SKILL.md to validate requirements before execution, which serves as a safeguard against malicious instructions in the input data.
Ingestion points: External work descriptions or plan files enter the context via the input_document argument in SKILL.md.
Boundary markers: The skill uses XML-style <input_document> tags to delimit untrusted input.
Capability inventory: The skill has the capability to execute shell commands (git, gh), modify files, and spawn subagents.
Sanitization: Initial triage (Phase 0) and clarification (Phase 1) steps are used to interpret requirements and confirm scope with the user.

ce-work