GPT-6.1 Sol and Codex Agentic Coding Prompt Pack: 50 Prompts

GPT-6.1 Sol costs $2.00 per million input tokens and $10.00 per million output, and every vague prompt burns that on retries. These 50 prompts each have one job, a stop condition, a verification step and a named output shape. They work in Codex, in a terminal agent, or with any capable coding model.

What You Get

• 50 prompts in 7 groups. Planning (8), refactor (8), tests (7), review (7), incident (7), migration (7) and docs (6), each with a use-when line and an expected-output description.
• Fill-in variables. Every prompt uses plain {{variable}} slots with a description for each, so you know exactly what context to paste.
• prompts.json. The full pack as structured data with id, title, when to use, prompt, variables and expected output, ready to load into your own tooling.
• Token budget table. Estimated prompt size, typical context, output and cost per run for every prompt on GPT-6.1 Sol and Claude Haiku 5.5, plus the assumptions behind each figure.
• Render script. A standard-library Python tool that fills variables from flags or files, warns on anything unset, and can fail in strict mode.
• Validator script. Checks that your edits keep ids unique and every variable declared and used.
• Usage guide. Seven design rules, proven prompt chains for features, bugs and upgrades, model choice by task, dated plan and allowance facts, and common failure modes.
• Safety clauses built in. Stop conditions, refusal clauses and scope fences that keep agents from patching symptoms or editing during a planning step.

How to Use

1. Pick the job from the README index, for example REV-01 for a correctness review.
2. Run render_prompt.py with your variables, or copy the block from the matching markdown file.
3. Paste it into Codex or your agent and read the first response before approving edits.
4. Chain prompts using the recipes in the usage guide.
5. Check prompt-budget.csv before you run a prompt over a large repo.
6. Edit the prompts for your stack and run validate_prompts.py to keep the JSON consistent.

Who This Is For

• Engineers using Codex, GPT-6.1 Sol or another coding agent every day.
• Tech leads who want repeatable prompts for reviews, incidents and migrations.
• Teams standardising how agents are briefed across repos.
• Solo developers who want fewer retries and a predictable token bill.

The prompt text is small, so the cost sits in your context. Use the budget table to plan it. Browse more in the AI prompt packs collection, or try AI Prompt Cost Calculator.