The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by design, as it involves an agent reading and processing untrusted data (source code and requirements) that is then interpolated into prompts for a second reviewer agent.
Ingestion points: Prompts in SKILL.md ingest {DIFF_CONTENT}, {CAVEKIT_CONTENT}, and {PLAN_CONTENT} from the local workspace.
Boundary markers: The provided templates use Markdown headers (e.g., '## Code changes:') to structure the prompt, though these do not prevent adversarial instructions within the data from influencing the model.
Capability inventory: The skill triggers shell commands and CLI tools (codex, scripts/codex-review.sh) to automate the review workflow.
Sanitization: No explicit validation or escaping of ingested file content is described in the prompt templates.
[COMMAND_EXECUTION]: The documentation references the execution of local shell scripts and CLI tools (e.g., git diff, scripts/codex-review.sh, and {LOOP_TOOL}) to facilitate the iterative build-and-review cycle.
[EXTERNAL_DOWNLOADS]: The skill references the installation of the @openai/codex package from the official NPM registry, which is a well-known service.

peer-review