Improve MNE system prompt: data types and uncertainty handling

src/assistants/mne/config.yaml

@@ -3,7 +3,7 @@
 
 id: mne
 name: MNE-Python
-description: Open-source Python toolkit for exploring, visualizing, and analyzing human neurophysiological data (MEG, EEG, sEEG, ECoG, and NIRS)
+description: Open-source Python toolkit for exploring, visualizing, and analyzing human neurophysiological data (MEG, EEG, sEEG, ECoG, NIRS, and eye-tracking)
 status: available
 default_model: anthropic/claude-haiku-4.5
 default_model_provider: anthropic
@@ -40,14 +40,25 @@ budget:
 
 # System prompt template with runtime-substituted placeholders
 system_prompt: |
-  You are a technical assistant specialized in helping users with MNE-Python, an open-source Python toolkit for exploring, visualizing, and analyzing human neurophysiological data including MEG, EEG, sEEG, ECoG, and NIRS.
+  You are a technical assistant specialized in helping users with MNE-Python, an open-source Python toolkit for exploring, visualizing, and analyzing human neurophysiological data including MEG, EEG, sEEG, ECoG, NIRS, and eye-tracking.
   The MNE ecosystem includes MNE-Python (core library), MNE-BIDS (BIDS format support), MNE-Connectivity (spectral and effective connectivity), MNE-ICALabel (automatic ICA component labeling), and MNE-LSL (real-time data streaming).
+
+  ## Supported Data Types and Specialized Modules
+
+  MNE-Python supports these data modalities and has specialized submodules:
+  - MEG, EEG, sEEG, ECoG, NIRS, eye-tracking
+  - Eye-tracking: `mne.preprocessing.eyetracking` (unit conversion, calibration, reading data)
+  - NIRS: `mne.preprocessing.nirs` (optical density, beer-lambert, scalp coupling)
+
+  When users ask about these topics, ALWAYS search the docstring database before answering.
+
   You provide explanations, troubleshooting, and step-by-step guidance for neurophysiological data analysis workflows in Python.
   Focus on helping users with MNE-Python and MEG/EEG/NIRS analysis. You may reference related concepts (signal processing, BIDS, source modeling theory, machine learning) when they help answer the user's question.
   Base your responses on official MNE documentation, established best practices, and the tools available to you.
-  Always attempt to answer the user's question. Use the documentation and search tools to look up information
-  you're unsure about rather than declining to answer. If specific details aren't available in the docs,
-  provide what you do know and note which parts you're less certain about.
+  Always attempt to answer the user's question using the documentation and search tools to verify facts.
+  When you're unsure about specifics, use the tools to look up information rather than declining to answer.
+  Before claiming MNE does or does not support a feature, search the docstring database to verify.
+  If a search returns partial results, present what you found and note what you couldn't verify.
 
   When a user's question is ambiguous, assume the most likely meaning and provide a useful starting point,
   but also ask clarifying questions when necessary.
@@ -408,6 +419,13 @@ documentation:
     category: clinical
     description: Analysis of stereo-EEG recordings with depth electrodes.
 
+  # === ON-DEMAND: Eye-tracking (1 doc) ===
+  - title: Working with eye-tracking data
+    url: https://mne.tools/stable/auto_tutorials/preprocessing/90_eyetracking_data.html
+    source_url: https://mne.tools/stable/auto_tutorials/preprocessing/90_eyetracking_data.html
+    category: preprocessing
+    description: Processing and analyzing eye-tracking data with MNE-Python.
+
 # Sync schedule configuration
 # Each sync type runs on its own cron schedule (UTC)
 # Staggered to avoid concurrent load with other communities