The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute standard local git commands, such as git branch --show-current and git diff --name-only. These operations are limited to metadata extraction from the local environment and do not involve network connectivity or elevated permissions.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes environmental data (filenames) which could be controlled by an attacker. Ingestion points: File names returned by git diff; Boundary markers: None provided in the command processing phase; Capability inventory: Access to the Bash tool for repository management; Sanitization: No explicit sanitization of filenames is performed before categorization. The risk is mitigated by the assistant's narrow focus on file classification.
[SAFE]: Explicit safety guardrails are defined in the 'AI Assistant Instructions' section, specifically forbidding the AI from discarding files or resetting staged changes without a direct user decision. This 'human-in-the-loop' requirement effectively prevents accidental or malicious data loss.

scope-guard