autoresearch-hypothesizer

To propose and rank research paths for an autonomous research loop, the skill reads task_plan.md and findings.md, then drafts 3–5 paths under the run's metric and budget, outputting results to a scratch directory.

Automated research replanning during a plateau or regression works by generating adjacent variants or divergent pivots based on the current session state, enabling the research loop to resume progress without manual intervention.

Yes, you can prioritize research paths by expected information value under a GPU budget. The skill ranks proposed paths based on the run's defined metric and resource constraints to enable efficient experimentation.

No, the autoresearch hypothesizer does not modify canonical task plan and findings files. It writes its proposed research paths to a scratch directory and notifies the orchestrator with a pointer, leaving canonical outputs untouched.

The best way to handle escalations in research automation without external data access is to propose variants internally. The skill reads current session context to generate divergent pivots or adjacent variants, guiding experimentation without external execution.

Yes, an orchestrator is needed to manage the output of automated research path planning. The skill writes its ranked paths to a scratch directory and notifies the orchestrator with a pointer, ensuring the autonomous loop continues correctly.