autoresearch-hypothesizer

What problem does it solve?

Proposes and prioritizes autoresearch paths for an autonomous research loop, guiding experimentation without performing executions or accessing external data.

Core Features & Use Cases

• Initial draft: reads current session context and drafts 3–5 ranked research paths under the run's metric and budget.
• Replan: when a worker hits plateau, regression, or escalation, generates adjacent variants or divergent pivots to resume progress.
• Prioritize: ranks paths by expected information value under a GPU budget, enabling efficient resource use.
• Output handling: writes its results to a scratch directory and notifies the orchestrator with a pointer; keeps canonical outputs untouched.

Quick Start

Run the autoresearch hypothesizer on the current session to generate the initial 3–5 ranked research paths.