The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands (whoami, git remote) to retrieve metadata such as organization slugs and repository names for session bootstrapping.
[COMMAND_EXECUTION]: The workflow includes a step where the agent is instructed to execute git commands using arguments (git_argv) provided dynamically by the prepare_review_diff tool.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes untrusted source code and unified diffs.
Ingestion points: Unified diff text and code snippets are passed as arguments to the check_changes and check_code tools.
Boundary markers: Not explicitly defined in the provided instructions.
Capability inventory: Shell execution (whoami, git), file reading, and MCP tool invocation.
Sanitization: Relies on the external MCP server's logic; however, the skill includes explicit instructions to stop operations if a 'misaligned' verdict is returned.

check-code