openclaw - 💡(How to fix) Fix pdf tool: MiniMax provider blocked from extraction fallback mode

Summary

The pdf tool excludes MiniMax providers (minimax and minimax-portal) from both native and extraction fallback modes, returning "Unknown model" even when valid MiniMax models are configured as pdfModel or imageModel.

Steps to reproduce

1. Configure MiniMax OAuth (minimax-portal) with M2.7 model
2. Set agents.defaults.pdfModel.primary = "minimax-portal/MiniMax-M2.7"
3. Call pdf tool with a local PDF file
4. Error: "Unknown model: minimax-portal/MiniMax-M2.7"

Root cause

In minimax-vlm-BcnwK449.js, isMinimaxVlmModel() check at line 74 explicitly throws "Unknown model" for MiniMax provider/model combinations. This hard block prevents the extraction fallback path from ever being reached.

The model resolution flow:

1. resolveProviderVisionModelFromConfig correctly finds MiniMax-VL-01
2. isMinimaxVlmModel blocks it before extraction fallback
3. Text-only M2.7 is not found by vision model search at all

Expected behavior

For non-VL-01 MiniMax models (like M2.7), or even for VL-01, the extraction fallback should work:

1. Extract PDF text via bundled pdfjs-dist
2. Send extracted text to the model
3. Skip image rendering for text-only models

Impact

Users with only MiniMax (Coding Plan/OAuth) auth cannot use the pdf tool, even though:

• Chat works (M2.7 via Anthropic-compatible)
• Image understanding works (VL-01 via VLM endpoint)
• Music generation works (music-2.6 via OAuth)
• Web search works (plugin)

Environment

• OpenClaw 2026.5.12 (f066dd2)
• MiniMax Coding Plan (OAuth, minimax-portal)
• macOS + webchat channel