Add optional autoresearch hint model tool

[0exec]

Summary

Adds an opt-in ask_autoresearch_hint tool that lets an autoresearch loop ask a configured larger Pi model for concise strategy advice when progress stalls.

The tool is disabled by default and side-effect free: it does not edit files, switch the active model, commit, revert, or write autoresearch.jsonl.

Behavior

Configured through autoresearch.config.json:

{
  "hints": {
    "enabled": true,
    "provider": "opencode-go",
    "model": "glm-5",
    "thinkingLevel": "low",
    "maxRecentRuns": 3,
    "maxCallsPerSession": 2,
    "timeoutSeconds": 120
  }
}

[davebcn87]

Interesting, I've been achieving this same as a separate extension, that I use across other tasks more than autoresearch, but didn't consider putting it as part of the core extension.

I wonder if this could be simplified, it feels a bit too complicated even to understand all config params.

Also have you used it? What are the results with it?

[0exec]

Interesting, I've been achieving this same as a separate extension, that I use across other tasks more than autoresearch, but didn't consider putting it as part of the core extension.

I wonder if this could be simplified, it feels a bit too complicated even to understand all config params.

Also have you used it? What are the results with it?

Tried a while ago, its outdated now but this was just more a draft for the potential feature.