“Did you even fix the automation???”
The agent repaired the report. It had not updated the automation prompt.
A Method is a reusable workflow file with instructions and checks for each step. Run it with new inputs, inspect the results, and save corrections for the next run.
I searched six months of my Codex history with 147,000 messages and these are 3 things I kept saying.
The agent repaired the report. It had not updated the automation prompt.
The agent stopped at a fixable failure. Its blocked status had no recorded reason.
The agent saved a debugging guide. Eight days later, I had to explain the procedure again.
Method keeps the instructions in a workflow file and each run’s results in a separate record. You can read what should happen, inspect what did happen, and review what changes next.
Bring a prompt, a conversation, or a screen recording. Include an example of the result you want.
Your coding agent turns that material into a Method: a file that defines the inputs, steps, instructions, and checks.
Read the steps before you run them. Ask your agent to change anything that is missing or wrong.
Give your Method the next topic, document, or task. The local executor runs its steps using the tools and connections available on your computer.
A research Method can take a new topic, collect sources, and write another brief. Each run keeps its own outputs and check results.
Try the research brief exampleOpen the run record to see which steps completed and which failed.
Select a step to read its instructions, output, and check results. Give that evidence to your coding agent so it can investigate the failure.
Ask your agent to update the Method. Review the changed instructions in the workflow file, then run it again.
Inspect the new results to see whether the repair worked. Earlier run records remain available for review.
Start on your own computer, with your own files, browser tools, and API connections.
When you need scheduled cloud runs or want to use a Method with your customers, work with us through our cloud pilot. We’ll help configure the tools, connections, access, and customer inputs the workflow needs.
Explore cloud deployment
A Method is a YAML file. You can read and edit its steps.
The open-source SDK executes those steps with Codex and saves the results. Inspect the schema and runner, or start with a complete research example.

A Method is a YAML file that defines a workflow’s inputs, steps, instructions, and checks.
The file describes the work to do. Each run produces a separate record of the outputs and check results.
Yes. Bring the prompt or conversation where you worked out the task, or a screen recording showing the process.
Your coding agent can use that material to build a Method.
Use Codex or Claude Code to help create and edit the Method and start a local run. The Method Executor runs on your computer; the open-source SDK uses Codex to execute steps.
Your Method dashboard shows the saved run records.
The run record identifies the failed step and saves its output and check results.
Ask your coding agent to investigate that evidence, update the Method, and run it again. You can review the instruction changes and inspect the next result.
Scheduled cloud runs are available through our pilot. Scheduling is not yet available from the dashboard.
Book a call. We’ll discuss the workflow you want to run, the access it needs, and pilot pricing.
Give your coding agent the work you want to repeat.
Run a research Method, read its steps, and adapt it to your topic.
Try the research brief exampleRun a Method in the cloud or use it with your customers.
Explore cloud deployment