The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands (git rev-parse --show-toplevel and mkdir) to determine the repository root and create the docs/rules/review/ directory for storing the generated reviewer files.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection because it incorporates user-supplied text directly into markdown files that guide the behavior of a secondary agent. 1. Ingestion points: User input is collected via AskUserQuestion for fields including 'Best for', 'Detection signals', 'Worked scenarios', and 'Checklist' items. 2. Boundary markers: The references/template.md file does not utilize boundary markers or explicit instructions to ignore embedded commands within the user-provided sections. 3. Capability inventory: The generated markdown files are loaded as behavioral instructions for the deep-review agent, which possesses file reading and shell capabilities. 4. Sanitization: While the 'name' field is validated for kebab-case format, other free-form text fields are interpolated into the final file without sanitization or escaping.

reviewer-creator