The Agent Skills Directory

Command Execution: The skill is designed to interact with the local file system and repository using standard Git commands such as git add -A and git commit. These operations are fundamental to the skill's purpose of automating the commit workflow.
Indirect Prompt Injection Surface: The skill processes information from the session history and repository diffs to generate commit messages and determine which files to stage. This creates a surface where the agent might interpret untrusted data (such as code comments or log files) to perform actions. The skill includes a manual verification step (Step 4) to mitigate the risk of staging unintended artifacts.
Ingestion points: Session history and outputs from git status, git diff, and git diff --staged (SKILL.md).
Boundary markers: None explicitly defined to separate untrusted code content from the agent's instructions.
Capability inventory: Execution of shell commands (git add, git commit) via the agent's environment.
Sanitization: Step 4 instructs the agent to 'sanity-check' and flag suspicious files (build artifacts, logs) to the user before proceeding.

commit