research-cycle

Automating a full autoresearch cycle requires a workflow that processes your hypothesis, executes director review, and deploys on Modal for evaluation. This Skill orchestrates that end-to-end sequence to eliminate manual steps and decision fatigue.

Testing an autoresearch workflow is best done using scout mode to validate loop inputs or dry-run mode to simulate execution without deploying to Modal, ensuring safe validation before committing resources.

Modal access is required for full execution and evaluation during an automated research cycle. You can use dry-run mode to simulate the workflow and validate your hypothesis without needing actual Modal deployment access.

Budget state prevents overruns by enforcing safe defaults and checking limits stored in .loop/state before execution. This ensures automated research cycles halt when allocated experiment resources are exhausted.

Manually triggering an autoresearch loop is supported for testing loops or enforcing periodic review. You can initiate the cycle in scout, confirm, or dry-run modes depending on your specific validation needs.