Skip to main content
A scenario is a reusable starting situation. Once your agent or a manual tool call has created useful data, you can save that data as an ordinary repository file and start a fresh environment from it.

​
In the inspector

  1. Start the environment with firedrill serve and use its tools.
  2. Open State & activity → Save as scenario.
  3. Give the scenario an ID and optionally a title. Preview the records.
  4. Choose Save new scenario. Firedrill creates <sourceRoot>/scenarios/<id>.scenario.json without overwriting anything.
The preview hides sensitive fields. If these exist, saving their original values requires explicit confirmation. The saved file is source, not an ignored report: inspect it for secrets before adding it to Git. The running world is unchanged. Start a new environment with the saved data:
You can also select the scenario in a drill’s scenarioId or with createLocalWorld({ scenario: "after-first-action" }).

​
From a test or coding agent

Both methods accept an optional title and packages array. With no package selection, all tools are captured. The other tools keep baseline data when only selected packages are captured. A stale preview, changed world baseline or tool code, existing scenario ID, external symlink, or invalid source produces an explicit error; nothing is overwritten. A reset invalidates the preview even if it happens to restore identical records.

​
What this captures

This is a tool-data seed, not a checkpoint. It includes current records and explicit deletions of records that otherwise would return from the baseline. It does not copy history, pending callbacks/events, consumed idempotency keys, random position, clock, actor permissions, or fault state. Those settings inherit the repository’s world baseline. Add explicit scenario settings in source if your next test needs them. Full runtime snapshots and data scenarios serve different purposes. The export is bounded to 50,000 records/deletions and 1 MiB of formatted JSON, matching the compiler’s source-file limit. Oversized exports fail before any file or directory is created. For a larger world, capture selected packages or create a scenario through ordinary source files. Exporting or saving data never marks an agent as tested or passed.