codex-research

Automate iterative research loops by initializing a workspace with a contract, then running guided or autonomous modes to repeatedly invoke Codex for hypotheses and experiments. The workflow orchestrates design, guided-loop, and autonomous-loop modes for bounded iterative work and overnight exploration.

A contract-and-state workflow enforces structured experimentation by maintaining program.md, contract.md, state_snapshot.md, and ledger.tsv to track hard gates, experiment status, and control actions. This ensures goal-oriented iterative work remains bounded and auditable throughout the research process.

Initialize a workspace, write a contract defining your goal boundaries, and start the guided loop to begin the first research round. The CLI workflow supports init, status checks, and running iterations with max rounds, optional web search, and full automation settings.

Yes, autonomous-loop mode supports overnight exploration by running Codex iterations without manual intervention. You can configure max rounds and enable full automation to let the system form hypotheses, test changes, and record results in the ledger autonomously.

Yes, the workflow records per-round data in a ledger.tsv file and a rounds/ directory containing prompts, messages, events, and evidence. This structured recording tracks experiment status and control actions across all iterations for later review.

Automated Codex research loops require a structured workspace with predefined contracts and are bounded by max rounds settings. The workflow enforces hard gates through state snapshots, meaning experiments cannot proceed beyond contract-defined boundaries without explicit control actions.