The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git CLI commands and user-defined verification scripts to manage the repository and confirm changes.
Evidence: The scoped-commit-executor.md subagent is responsible for staging files, running git commit, and executing a VERIFICATION_HINT (such as npm test) to ensure commit quality.
[EXTERNAL_DOWNLOADS]: The skill fetches technical documentation and stylistic guidelines from external URLs to assist with commit planning and Git mechanics.
Evidence: references/external-sources.md contains a list of URLs including official Git documentation (git-scm.com), Anthropic's engineering blog, and industry standard sites like conventionalcommits.org.
Security Control: The skill implements a progressive disclosure policy where subagents fetch these URLs and return only a one-line summary to the orchestrator, preventing potential prompt injection from external content into the main agent context.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by processing untrusted data from the repository being managed.
Ingestion points: scoped-state-summarizer.md reads Git diffs and local documentation files (via CONTEXT_LOCATION).
Boundary markers: The skill does not explicitly use delimiters for repo content, but it uses a summary-first approach to avoid passing raw patches into the main context.
Capability inventory: The skill can execute Git commands and verification scripts via the scoped-commit-executor.md subagent.
Sanitization: The subagents are instructed to summarize behavioral changes rather than returning raw hunks or document text.

committing-scoped-changes