The Agent Skills Directory

[COMMAND_EXECUTION]: Employs local git commands for repository inspection and staging, and utilizes tools such as commitlint, gh, or glab for extended functionality.
[PROMPT_INJECTION]: Analyzes untrusted data from the local repository (e.g., via git diff), creating a surface for indirect prompt injection. Ingestion points: File content retrieved through git commands. Boundary markers: None. Capability inventory: git add, git commit. Sanitization: Mandatory user approval before committing and specific instructions to identify and exclude sensitive information.
[DATA_EXFILTRATION]: Reads local file content to generate commit messages. The skill proactively mitigates exposure risks by instructing the agent to detect secrets, credentials, and local configuration files, ensuring they are not staged without explicit user direction.

code-commit