The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard Git commands (git status, git diff, git log, git commit) to manage the repository's working tree. It employs secure shell practices by using git commit -F - with heredocs to pass commit messages, avoiding potential shell injection issues associated with the -m flag.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local environment, specifically file diffs and historical commit logs, to summarize work. The risk is mitigated by the disable-model-invocation: true setting, which ensures the skill only runs upon explicit user command (/commit-all), and by the requirement for user confirmation before committing.
[DATA_EXPOSURE]: The workflow explicitly instructs the agent to identify and exclude untracked files that may contain secrets or sensitive information (e.g., one-off scripts, files omitted from .gitignore) rather than staging them automatically.

commit-all