The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses the security find-generic-password command to programmatically retrieve an OAuth token from the macOS Keychain. While this is used to authenticate the claude CLI, programmatic access to system credential stores is a sensitive operation.- [COMMAND_EXECUTION]: The script executes several shell commands, including security for credential retrieval and the claude CLI tool to process review requests.- [PROMPT_INJECTION]: The skill processes user-provided questions and file paths which are then passed to a secondary agent. This creates a surface for indirect prompt injection where malicious content in the referenced files could attempt to influence the reviewer's judgment.
Ingestion points: Processes a user-defined question and file content from specified paths provided in the bash script heredoc.
Boundary markers: Uses a shell heredoc (EOF) to group the prompt, but lacks explicit boundary markers or instructions to the sub-agent to treat embedded file content as untrusted data.
Capability inventory: The skill executes shell commands and has read access to the system Keychain.
Sanitization: No explicit sanitization or validation is performed on the user prompt or the contents of the files being analyzed before they are passed to the CLI tool.

taste-review