The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local git commands (git status, git add, git diff, git ci) to manage the repository state and create commits.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external data from git diff output to generate commit messages. This is a potential surface for indirect prompt injection if a file being committed contains malicious instructions intended to influence the LLM's commit message generation.
Ingestion points: The skill reads untrusted data via git diff --cached and git diff --cached --stat in step 2.
Boundary markers: None are specified for the diff content being processed.
Capability inventory: The skill has capabilities to stage files (git add), write files (Write tool), and execute commit commands (git ci).
Sanitization: The skill mitigates this risk by explicitly requiring the user to review and approve the generated commit message in step 4a before the final commit command is executed.
[COMMAND_EXECUTION]: The skill uses gitflow-toolkit, which is a tool provided by the author. It includes an instruction to run gitflow-toolkit install if the tool is not found. This is a standard setup instruction for the skill's intended environment.

commit