OpenAI¶
What the openai/default files change, laid over the default rules for every openai: model. Every rule not listed here reads as it does by default.
wording¶
Released 2026-10-07; a pin reads it as wording = "1.0.0" under [lint.rules].
Writing for openai/default
Over-triggering on emphasis is documented for GPT-4.1 and GPT-5; OpenAI's GPT-5.2 guide still uses 'You MUST'.
- On GPT-4.1 and GPT-5, write requirements plainly instead of in capitals.
- The vendors disagree on emphasis: decide by an experiment on your route, not by the rule alone.
Sources: OpenAI, GPT-4.1 prompting guide; OpenAI cookbook, GPT-5 prompt optimizer.
From openai/default, 1.0.0.
| Rule | Finds | Default | On openai: models |
|---|---|---|---|
wording/capitals |
Emphasis in capitals | info | warn on gpt-4.1*, gpt-5; info on the rest |
wording/exclamation-marks |
Repeated exclamation marks | info | warn on gpt-4.1*, gpt-5; info on the rest |
instructions¶
Released 2026-10-07; a pin reads it as instructions = "1.0.0" under [lint.rules].
Writing for openai/default
GPT-4.1 and GPT-5 read hedges literally.
- Hedges ('try to', 'if possible') are read literally by GPT-4.1 and GPT-5: make each firm or plainly optional.
Sources: OpenAI, GPT-4.1 prompting guide; OpenAI cookbook, GPT-5 prompt optimizer.
From openai/default, 1.0.0.
No rule changes in this release.
tools¶
Released 2026-10-07; a pin reads it as tools = "1.0.0" under [lint.rules].
Writing for openai/default
For OpenAI reasoning models, examples inside tool descriptions may hurt tool selection.
- Leave examples out of tool descriptions; describe what the tool does, returns, and when not to use it.
Sources: Claude API docs: Define tools; OpenAI, Function calling guide; Anthropic skills repository, prompt-audit guide.
From openai/default, 1.0.0.
| Rule | Finds | Default | On openai: models |
|---|---|---|---|
tools/examples-in-tool |
Examples inside a tool description | warn | own fix |