The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the analysis of code diffs.
Ingestion points: Untrusted data from code changes is read via git diff and git diff --cached in the execution steps.
Boundary markers: No delimiters are used to isolate the diff data from the agent's instructions.
Capability inventory: The skill can perform file staging with git add -A and create commits with git commit -s.
Sanitization: No filtering or validation of the code content is performed before processing.
Remediation: Wrap the code diff output in unique delimiters and explicitly instruct the agent to ignore any natural language instructions found within the code changes.
[COMMAND_EXECUTION]: The skill executes local shell commands to manage the git repository.
Evidence: The execution steps include calls to git status, git diff, git add -A, and git commit -s. These are consistent with the skill's primary purpose.

commit-msg