The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to interact with the Git version control system.
Evidence: Use of git status, git diff, git diff --staged, git log, and git commit to analyze changes and record history.
[INDIRECT_PROMPT_INJECTION]: The skill processes output from git diff and git status, which are external data sources that could contain adversarial content designed to influence the generated commit message or agent behavior.
Ingestion points: The agent is instructed to run and read the output of git diff and git status to ground its messages (SKILL.md, references/grounding.md, references/workflow.md).
Boundary markers: The skill mandates that the agent must "obtain explicit user approval" of the proposed message before running the final git commit command.
Capability inventory: The agent has the capability to write to the repository history via git commit and potentially perform destructive operations like git reset if requested by the user.
Sanitization: Safety rules explicitly instruct the agent to detect and warn the user about secret-bearing files like .env before adding them to a commit (references/safety.md).

commit-message