openai-api-toolkit
An Agent Skill for writing, reviewing, debugging and migrating code that calls the OpenAI API.
Its model tables and request linter were built by sending 55 request shapes (Responses API and
Chat Completions) to each of 29 OpenAI text models and recording what the API accepted. The
linter's verdict matched the API on all 1,594 recorded outcomes.
Not affiliated with OpenAI. "OpenAI" and "GPT" are trademarks of OpenAI.
Contents
| Path | Purpose |
|---|
SKILL.md | When to use the skill, the working loop, the rules that cause most failures |
references/models.md | Ids, context windows, output caps, prices (generated) |
references/request-rules.md | What each model accepts on each API, with exact error texts (generated) |
references/requests.md | Responses anatomy, streaming, structured outputs, files, embeddings, batches |
references/tools-and-agents.md | Function calling, the loop, hosted tools, the Agents SDK |
references/state-caching-cost.md | Conversation state, prompt caching, cost levers |
references/errors.md | Status codes and observed rejections with fixes |
scripts/openai-api.mjs | Offline CLI: models, lint, fix, convert, migrate, cost, compare, scaffold, guide |
The scripts use only the Node.js standard library, open no network connection and need no key.
Install
skills install @spboucher/openai-api-toolkit --client claude-code
Measured effect
See CHANGELOG.md for a with/without-skill evaluation against the live API, with its caveats.
Limits
A snapshot taken on the date in SKILL.md. Only text models visible to the probing key were
probed. Computer use, Realtime, fine-tuning, Assistants and Azure OpenAI were not probed. Cost
output is an estimate. When the API and this skill disagree, the API is right.