The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard git commands (including git branch, git status, git add, and git commit) and invokes local project validation tools like npm and poetry. These operations are appropriate for a commit automation tool and are executed within the local repository context.
[DATA_EXFILTRATION]: The skill reads local agent activity logs from .claude/agents/activity/{branch}.jsonl. This data is used exclusively to generate co-author attribution trailers in the commit message and is not exfiltrated to any external domain.
[PROMPT_INJECTION]: The skill instructions do not contain any attempts to bypass safety filters or override agent behavior. The instructional text enforces git best practices, such as branch protection, using standard command-line logic.
[EXTERNAL_DOWNLOADS]: The skill does not perform any external network requests or downloads; all operations use local system tools or tools provided by the plugin environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data from activity logs. 1. Ingestion point: .claude/agents/activity/{branch}.jsonl. 2. Boundary markers: Absent. 3. Capability inventory: Git and shell commands. 4. Sanitization: Absent. The risk is considered low as the data is only used for commit message generation.

commit