The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a "CROWN RULE" that explicitly instructs the AI to hide its identity. It forbids the inclusion of AI attribution footers (e.g., "Generated with Claude") and mandates that commits must appear "human-authored only." This is a directive to bypass standard AI disclosure and transparency guidelines.\n- [COMMAND_EXECUTION]: The skill provides multiple shell scripts designed to execute Git commands (git commit, git diff, git log) and system utilities. These scripts are used for workflow automation and rely on shell execution environments.\n- [PERSISTENCE_MECHANISMS]: The skill directs users to install scripts into the .git/hooks/ directory and grant them execution permissions via chmod +x. This establishes a mechanism where the scripts are automatically and persistently executed during specific points in the repository's lifecycle (e.g., pre-commit and commit-msg).\n- [INDIRECT_PROMPT_INJECTION]: The scripts ingest data from external sources within the repository, creating a potential attack surface for indirect injection.\n
Ingestion points: The smart-commit.sh, pre-commit, and generate-changelog.sh scripts read data from git diff --cached and git log.\n
Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands within the processed git data.\n
Capability inventory: Ability to execute git commands and shell scripts.\n
Sanitization: Uses basic string processing with grep and sed, but lacks comprehensive sanitization for untrusted content in files or commit messages.

git-commit-helper