The Agent Skills Directory

[SAFE]: The skill is purely instructional and does not contain any executable scripts, tools, or binary components. It follows a secure design by deferring all repository interactions, authorization checks, and execution to a separate 'git-delivery' skill.
[INDIRECT_PROMPT_INJECTION]: The skill identifies an ingestion point at references/local-conventions.md (SKILL.md) for project-specific rules. It includes strong boundary markers that explicitly forbid these local conventions from overriding user boundaries, repository facts, or the primary authorization policy. The skill relies on external capabilities via the 'git-delivery' sibling skill for all sensitive operations, and while no specific data sanitization is mentioned, the instructional constraints effectively mitigate the risk.

git-workflow