The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external CLI tools (codex and cursor-agent) via shell commands. It uses command substitution $(cat "$review_prompt_file") to pass generated instructions as arguments, a pattern that requires robust shell handling to prevent unintended execution if file contents are manipulated.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as source code and git changes, by passing them to a secondary AI agent for review. This exposes the sub-agent to potential instructions embedded within the reviewed material.
Ingestion points: The skill accepts git references and file paths as arguments, which are then read into the agent's context.
Boundary markers: The instructions include a specific prompt template designed to constrain the sub-agent's response format and scope.
Capability inventory: The skill can execute shell commands and write report files to the local file system.
Sanitization: While the codex command includes a --sandbox read-only flag, no such sandboxing is explicitly configured for the cursor-agent tool in the provided instructions.- [DYNAMIC_EXECUTION]: The skill dynamically constructs and executes shell commands at runtime based on the content of temporary files, representing a form of dynamic command assembly.

agent-review