The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions proactively address command injection risks by mandating that the agent pass commit messages as data (e.g., using git commit -F <file>) rather than through shell interpolation.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests potentially untrusted data from repository diffs and rules, it includes strict constraints to mitigate injection attacks: it forbids inventing intent, restricts message formats to a rigid schema, and requires the agent to stop and wait for the user if unrelated concerns are staged together.
[SAFE]: The skill follows the principle of least privilege by requiring explicit user authorization for each commit and refusing to perform amendments or push operations.

hl-commit