The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as source code, test files, and requirement documents which are ingested using Read, cat, and grep. This creates a surface for indirect prompt injection if these files contain malicious instructions disguised as comments or data.
Ingestion points: TEST_FILE, SOURCE_FILE, and request documents located via docs/features/*/requests/*.md are read into the agent's context.
Boundary markers: Prompt templates in references/ use Markdown headers and variable placeholders to structure the input, providing some separation between data and instructions.
Capability inventory: The skill possesses the ability to write files (Write tool), execute local scripts via node, and delegate work to other agent tasks (Task tool).
Sanitization: There is no evidence of sanitization or filtering of the content read from these external files before it is interpolated into the prompts for the Codex reviewer.
[DYNAMIC_EXECUTION]: The skill generates unit test code based on project templates and saves them to the file system. It also executes local Node.js scripts (scripts/lib/review-dispatch.js and scripts/validate-family-sentinel.js) to manage its internal logic and validate results.
[COMMAND_EXECUTION]: The skill uses Bash tools to execute git, node, grep, and ls commands to analyze project structure, search for evidence, and run validation checks.

test-review