The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a runner script using the node engine to manage the debate lifecycle, including starting, polling, and stopping processes as described in references/workflow.md.
[EXTERNAL_DOWNLOADS]: The documentation requires a dependency hosted on GitHub (github:lploc94/codex_skill) to be installed via npx, which involves downloading code from the author's repository.
[REMOTE_CODE_EXECUTION]: The recommendation to use npx github:lploc94/codex_skill facilitates the execution of remote code directly from a GitHub repository during the setup phase.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It processes untrusted data from local plan files and user requests by interpolating them into internal prompt templates used for implementation review.
Ingestion points: The contents of plan.md and the USER_REQUEST string are used to populate variables in the prompt templates found in references/prompts.md.
Boundary markers: The templates use standard Markdown headers (e.g., ## Plan Location) to separate sections, which provides minimal protection against adversarial content within the ingested files.
Capability inventory: The skill possesses capabilities to execute system commands through the node-based runner and to modify local files when applying plan edits.
Sanitization: There is no explicit evidence of input validation, sanitization, or escaping of the content being interpolated into the prompts.

codex-plan-review