#Observe, answer, and cancel Runs

Create a Run once, retain its ID, and use that ID for every subsequent operation.

intloom run intent --prompt-file ./request.txt --json
intloom runs --loom intent
intloom attach RUN-example --json
intloom wait RUN-example --timeout 30000
intloom watch RUN-example

wait defaults to action_or_terminal. --until terminal waits past questions without answering them. Timeouts and Ctrl+C end observation only. watch reconciles snapshots on reconnect without resubmitting execution.

#Answer the current action

Read pendingAction and retain its exact ID. A questions action requires every question exactly once:

[
  {"questionId": "platform", "isSkipped": false, "answer": "Web"},
  {"questionId": "notes", "isSkipped": true}
]

Use actual question IDs. Skip only questions that allow skipping. A confirmation action uses:

{"isConfirmed": false, "feedback": "Please revise the storage requirement."}

intloom answer RUN-example --action ASK-example --answer-file ./answer.json

Stale or invalid answers preserve the pending action. After an uncertain response, inspect the Run before submitting again.

#Cancellation and uncertain writes

intloom cancel RUN-example
intloom operation OPERATION-UUID --instance INSTANCE-ID --json

Cancellation becomes failed with RUN_STOPPED and preserves committed data and project effects. Query uncertain admission receipts using the original operation and instance. Explicit replay requires the original parameters, --operation-id and --instance.

Run statuses and receipt outcomes describe different things; see CMD reference. Resuming a restored mirror is a separate operation described in mirrors.