The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to the combination of reading untrusted external files and carrying out write actions.
Ingestion points: Ingests untrusted repository data through commands like git diff, git diff --cached, git status --short, and git log within SKILL.md.
Boundary markers: Lacks explicit boundary delimiters or structural markers to distinguish raw source code modifications from prompt instructions.
Capability inventory: Has extensive tool capabilities to execute commands including git add, git commit, git push, and git filter-branch as detailed in SKILL.md.
Sanitization: Features a safety check that filters out credentials and .env files from commits, but does not escape or sanitize general code contents or commit logs against linguistic prompt injection.
[COMMAND_EXECUTION]: Executes shell and git operations directly on the user's host environment, including invoking local python scripts via git filter-branch and performing destructive actions such as git push --force when the explicit force keyword is specified.

git-commit