The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell-level git commands, including git add ., git diff, git commit, and git push. This involves broad filesystem access and the execution of external binaries.
[DATA_EXFILTRATION]: The skill is designed to perform git push, which transmits the local codebase to remote repositories. While this is the intended purpose, it represents a path for data to leave the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data through git diff.
Ingestion points: Reads code changes from the workspace via git diff --cached.
Boundary markers: None. The agent analyzes the raw diff content directly.
Capability inventory: Includes git commit (write operation) and git push (network operation).
Sanitization: There are no instructions to sanitize or escape the content of the diff before it is used to generate the commit message. A malicious user could include instructions inside code comments that influence the generated commit message or attempt to escape the git commit -m command.

semantic-git-commit