---
name: codex-prompting
description: >-
  Use when improving Codex-style task prompts, AGENTS.md instructions, review
  briefs, subagent handoffs, or implementation requests so they are precise,
  scoped, testable, and safe for coding agents.
version: "1.0.0"
triggers:
  - improve this codex prompt
  - write agent instructions
  - draft an implementation brief
  - make this prompt testable
negative_triggers:
  - write application code only
  - summarize a meeting transcript
allowed_tools: []
required_tools: []
trust_level: local
safety_class: read_only
examples:
  - "Rewrite a coding-agent task prompt with scope, constraints, acceptance criteria, and verification commands."
tests:
  - "sb skills validate --path skills --strict"
performance:
  notes: "Best for precise coding-agent prompts, handoffs, and repo instructions."
---

# Codex Prompting

## Purpose

Create prompts and instruction files that coding agents can execute without
guesswork.

## Instructions

1. State the objective, owned scope, and out-of-scope areas.
2. Name the exact files, registries, APIs, or commands the agent should inspect first.
3. Include constraints for safety, permissions, style, tests, and working-tree hygiene.
4. Add acceptance criteria that can be verified locally.
5. Prefer concrete commands over vague quality goals.
6. Keep the final prompt direct enough to paste into a coding-agent session.

## Checklist

- Goal and reason are explicit.
- Ownership boundaries are explicit.
- Required discovery steps are listed.
- Validation commands and expected outcomes are listed.
- Final handoff format is specified.