| name | nice-friend |
| description | Consult OpenAI Codex as a trusted friend for design review, architecture feedback, or technical advice. Use when the user asks to consult Codex or wants a second opinion. |
| argument-hint | ["consultation topic or question"] |
Nice Friend - Codex Design Consultation
Consult OpenAI Codex CLI (codex exec) as a trusted, knowledgeable friend for design review and technical feedback.
How to use
- Prepare context for Codex by reading relevant files (DESIGN.md, source code, etc.)
- Construct a prompt that includes:
- The project context (what trajix is, current architecture)
- The specific question or area to review
- A request for honest, constructive feedback as a friend would give
- Run
codex exec via Bash with the prepared prompt and relevant context
- Present Codex's feedback to the user
Prompt construction
When calling codex exec, use this pattern:
codex exec -s read-only "
You are a trusted friend and experienced software engineer reviewing a project.
Be honest, constructive, and specific. Point out both strengths and concerns.
If you see potential issues, suggest alternatives.
Project: [brief description]
Context: [relevant design/code excerpts]
Question: [the user's specific question or $ARGUMENTS]
"
Guidelines
- Always pass relevant project files as context (pipe file contents or reference them)
- Use
-s read-only sandbox mode since this is a consultation, not code generation
- Use
-C to set the working directory to the project root so Codex can read files
- If the topic is broad (e.g., "review the design"), focus on architecture, trade-offs, and potential pitfalls
- Present Codex's response clearly, noting it as Codex's perspective
- If Codex raises valid concerns, discuss whether to act on them