Agent is blocked
Work out why an agent stopped and get it moving again in the same session.
When an agent stops, Alan says why: in the status line of the turn, in a notice under the transcript, or in a question card. Read that first. Use the matching action below to continue in the same session. If the message asks you to start a new session, follow that instruction.
What Alan shows and what to do
| You see | Why | What to do |
|---|---|---|
| Waiting for input on the turn, and a card marked Waiting for your answer | The agent asked a question or needs a decision | Answer in the card. The turn continues. |
| Agent stopped · your answer starts a new run on a question card | The run ended before you answered | Answer anyway; your answer starts a new run with the context. |
| Run stopped | The run was stopped | Choose Continue to pick up where it left off, or send a new message. |
| The agent stopped before this run finished | The run ended before it finished | Choose Continue. |
| Waiting for (computer) to come online | The session runs on a computer that is offline. "Your message runs as soon as it reconnects." | Open the Alan desktop app on that computer, or wait. Alan stops waiting after a while and tells you how long is left. |
| Sign in to (agent) on (computer) | The coding agent CLI on that computer is signed out | "Open a terminal on that computer, sign in to the CLI, then send your message again." |
| The API key for (agent) was rejected | The provider refused the key | "Update it in settings, then send your message again." |
| (Agent) usage limit reached | The provider's plan limit is used up | Send again after the reset time shown, or switch to another model in the composer. |
| Stopped before it started | The turn was stopped or failed before the agent began | Send the message again. If it repeats, check the computer; see Runtime unavailable. |
| Not sent. The computer was stopped under your message | The message was cancelled before delivery | Start or wake the computer, then send it again. |
Turn states and message states are explained in full in Session lifecycle.
When the agent is stuck on the task, not the tooling
Sometimes the run finishes but the agent reports it can't go further: it lacks access to a repository, the instruction is ambiguous, or it is working in the wrong task or repository.
- Read the agent's last message and the task's latest activity or artifact. Look for the specific thing it needs.
- Give exactly that: the missing detail, a decision, or access. For repository access, see GitHub access issues.
- Reply in the same session so the agent keeps its context. A bare "try again" rarely helps unless the failure was temporary.
- If the task changed direction, start a new session on the task so the transcript stays focused.