OpenRouter's core value is model-agnostic access — users switch between GPT-4, Claude, Llama, Mistral without changing their app. But documentation and system prompts often break when switching because smaller models handle verbose docs differently than larger ones.
HADS (Human-AI Document Standard) solves this: https://github.com/catcam/hads
A lightweight Markdown tagging convention that works from GPT-4 down to Llama 3 8B:
## API Authentication
**[SPEC]**
- Method: Bearer token
- Header: `Authorization: Bearer <token>`
- Expiry: 3600s
**[NOTE]**
Switched from cookie auth in v2. Legacy docs mentioning cookies are outdated.
**[BUG] Token silently rejected after password change**
Symptom: 401 identical to expired token
Cause: All tokens invalidated on password change
Fix: Re-authenticate after any account operation
The AI manifest at the top tells any model what to read and skip — explicit instruction beats structural inference, especially for 7B models.
Why relevant to OpenRouter users:
- System prompts written in HADS work consistently across model tiers
- Context docs in HADS format reduce hallucination when switching to smaller/cheaper models
- Zero overhead — pure Markdown, no tooling
Would this be worth a mention in the docs or examples as a best practice for portable system prompts?
OpenRouter's core value is model-agnostic access — users switch between GPT-4, Claude, Llama, Mistral without changing their app. But documentation and system prompts often break when switching because smaller models handle verbose docs differently than larger ones.
HADS (Human-AI Document Standard) solves this: https://github.com/catcam/hads
A lightweight Markdown tagging convention that works from GPT-4 down to Llama 3 8B:
The AI manifest at the top tells any model what to read and skip — explicit instruction beats structural inference, especially for 7B models.
Why relevant to OpenRouter users:
Would this be worth a mention in the docs or examples as a best practice for portable system prompts?