The Agent Skills Directory

[SAFE]: The skill uses standard Git commands (status, diff, log, add, commit) to manage local version control. It includes safety-focused instructions such as avoiding bulk add commands in favor of explicit path staging and forbidding the use of --no-verify to ensure repository-required checks are always performed.
[SAFE]: The workflow explicitly denies unauthorized network activity, stating that the agent must not run git push, git pull, git fetch, or git merge. This prevents data exfiltration or unintended remote code ingestion during the checkpointing process.
[SAFE]: The skill implements a verification step before each commit, requiring the agent to run relevant tests and repository checks. While this involves executing code within the user's repository, it is consistent with the intended purpose of a development tool and relies on existing repository infrastructure.
[SAFE]: The skill addresses potential data exposure by referencing a shared workflow for secret detection and explicitly naming .env.example as the only allowed exception for public configuration templates.
[SAFE]: The skill manages a potential surface for indirect prompt injection where repository data is processed. This is handled safely through human-in-the-loop checkpoints (ingestion point: repository files; boundary marker: presenting plans to the user; capability inventory: local Git and test runners; sanitization: manual review of diffs).

git-commit-realtime