The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides guidance on executing git add and git commit commands. It includes instructions for managing index.lock contention by separating commands or using short delays, which is a standard operational practice for version control agents.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze code changes (potentially untrusted data) to generate descriptive commit messages. While this creates an ingestion surface, the risk is inherent to the skill's purpose.
Ingestion points: The agent reads the project's diffs and source code to derive commit summaries (SKILL.md).
Boundary markers: None provided to delimit the untrusted code content from the formatting instructions.
Capability inventory: The agent has the ability to stage and commit files via the git command line.
Sanitization: No specific filtering or validation of the input code changes is described.

git-commit