Come back to the same workspace#
Follow-ups reuse the files and conversation state. Idle compute can stop while the workspace remains. Continue the same task or Slack thread when you want to pick up where it left off.
Retained storage still costs money while compute is idle.
Cancel, close or remove#
| Action | What it does | What to remember |
|---|---|---|
| Stop a task | Stops the current work. | In a Slack workflow thread, only the owner can stop, with @AgentX stop. The workspace and changes stay. |
| Close a workspace | Checks for unpublished work before releasing workspace storage. | Preserve unfinished changes if closure refuses. |
| Remove the environment | A separate operator teardown procedure. | Use the release-matched instructions. |
Each person has a limit of open tasks (3 by default). Closing a task, or a workflow task finishing, frees its place. See the workspace lifecycle →
Understand task status#
Coding-tool tasks can be STARTING, RUNNING, SUCCEEDED, FAILED, CANCELLED, INTERRUPTED or CLOSED. CLOSED also covers a workflow task that finished; its workflow stage says whether its PRs merged (MERGED) or its plan was rejected (CLOSED).
SUCCEEDED describes task execution, not a guarantee that every check passed. A timed-out wait can leave the task running. Read its current status and evidence before retrying.
Understand the next action#
| What you see | What to do |
|---|---|
| A Quick/Full suggestion card | Press Start, switch path, or Just answer. Only the requester can choose. |
| Waiting for a plan decision | Read the current document, then approve or request changes in Slack or on the task page. |
| Checks or reviews are running | Nothing; a draft PR opens when both reviews pass. |
| A check or review blocked | Read the problems; send back to coding, retry reviews, or close the task. |
| The draft PR didn’t open | Retry opening the pull request, or close the task. |
| PR feedback is ready | Open details and choose the response. |
| Waiting for merge | Use GitHub review and merge controls. |
| Closeout retry needed | Check preserved artifacts and retry the failed cleanup. |
Task execution status is distinct from workflow stage and PR merge state.
What you pay for#
You pay for AWS services, retained storage and model usage. Task usage telemetry, installer estimates and budget alerts help explain usage. Budget alerts are notifications, not spending caps. Tasks of developers who connected a ChatGPT plan use that plan instead of the installation’s model where it can.
Assumes us-east-1, default Bedrock models, 1,000 turns, 100 worker sessions, 60 instance-hours and 10 retained workspaces. The source pricing calculation returns $211.13. This is an example, not a subscription price; actual usage and prices vary.
Check environment health#
With an installed CLI and operator access, use:
npx @preplabsai/rovara-code --env <env> doctor --region <region>
Read the reported failures before making changes. On mainline, doctor also reports Sign in with ChatGPT when this computer holds the admin sign-in. Upgrade and teardown instructions →
Task history and Canvas cleanup#
At terminal closeout, Rovara Code verifies its canonical artifacts and decisions and saves their references before deleting task Canvas copies, including earlier plan versions. A preservation or cleanup failure remains visible and retryable.
Canvas cleanup does not delete Slack thread messages or original GitHub comments. Artifact retention follows your configured policy; do not assume indefinite storage. Workspace storage and deployment teardown are separate operations.