The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using the user-supplied ref argument (e.g., diffity <ref> --no-open --new). Without explicit validation, this allows for command injection if a crafted ref string is provided.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to install the diffity CLI tool globally using npm install -g diffity if it is not already present on the system.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions found in CLAUDE.md files within the target repository. This creates a surface for indirect prompt injection where a malicious actor could influence the agent's review logic or behavior by placing instructions in these files.
Ingestion points: CLAUDE.md files in the repository root and subdirectories.
Boundary markers: None specified; the agent is instructed to read these files and follow the rules within.
Capability inventory: Execution of diffity and git commands, file reading, and potential network access via browser opening.
Sanitization: No sanitization or filtering of the content from CLAUDE.md is performed.

diffity-review