The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes repository content via git diff to generate commit messages. If a file being committed contains malicious instructions, it could influence the agent's output.
Ingestion points: Repository data is ingested via git diff and git status in handlers/commit-with-preview.md.
Boundary markers: The skill requires the user to review a preview file at ~/.copilot/session-state/{session_id}/commit_preview.txt and provide explicit confirmation via the ask_user tool.
Capability inventory: The skill has access to the bash tool to execute git operations like add and commit.
Sanitization: There is no specific evidence of sanitization to prevent the model from interpreting instructions embedded within the code changes it analyzes.
[COMMAND_EXECUTION]: The skill utilizes the bash tool to interact with the project's git repository. While this is the primary purpose of the skill, the execution of shell commands based on repository state is a significant capability.

git-conventional-commits