The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary shell commands defined in the Verify: sections of the input plan file (plan.md) using timeout 30 <cmd>. This is a core feature for verification but allows execution of instructions from the processed plan data.\n- [COMMAND_EXECUTION]: It executes a JavaScript companion script from a plugin cache directory: node "$CODEX_COMPANION" adversarial-review. This resource is associated with the vendor's environment.\n- [COMMAND_EXECUTION]: The skill performs local network operations via curl to a local API (localhost:41777) to manage code review annotations.\n- [COMMAND_EXECUTION]: It utilizes various standard local development tools such as pytest, ruff, eslint, and pyright, which involve subprocess execution.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Surface (Plan File): The skill ingests data from an external plan file and executes commands derived from its content.\n
Ingestion points: Reads the plan file at <path/to/plan.md> to extract verification commands.\n
Boundary markers: None; verification commands are executed directly in a shell environment.\n
Capability inventory: Arbitrary command execution, file system modification (rm, cp), and local network operations.\n
Sanitization: None; the skill does not validate or sanitize commands extracted from the plan file before execution.

spec-verify