Skip to main content
Use this page when an agent does not reply, a run failed, or a mission stopped moving. Start at the Health screen, then find the message that you see and follow its fix.

​
Start at the Health screen

The Health screen collects today’s failed runs in one place. The other screens show only a short link to it:
  • Home: “[n] runs today may need a look; details in Health”
  • Missions: “[agent] hit a snag today; details in Health”
  • An agent’s profile: “[n] runs today may need a look; details in Health”, or “Some runs today were interrupted by maintenance; details in Health”
  1. Select one of these links. The Health screen opens.
  2. Read the sections on the screen.
For more about Health and Runs, see See what your agents did.

​
Read the run detail

1

Open the run list

On a card in NEEDS A LOOK, select View in Runs →.
2

Open the run

Select the run. The run detail opens.
3

Read the label

Read the box at the top.
Open Technical details for the full error text. Give this text to support if you ask for help.

​
Error texts in a run

The run detail shows a short sentence. The raw text is under Technical details. These sentences are common:

​
Chat failure messages

When an agent cannot answer in chat, the chat shows one of these messages under your message. A message that the agent sent on its own, for example from a scheduled task, can show the same sentences when its run fails. It has no Retry, because there is no message of yours to send again. Ask the agent to try again.

​
An expired AI connection

When your AI connection expires, the dashboard tells you in more than one place:
  • The chat shows “Your AI connection expired. Replies will fail until you reconnect.” with Reconnect →.
  • Missions can show “Nothing ran today; your AI connection expired.” with Reconnect in AI Tokens →.
  • On AI Tokens, the account shows a status such as EXPIRED or LOGIN INVALID.
Fix: open Settings > AI Tokens and reconnect the account. Runs that use an expired account fail until you reconnect it. Squad does not retry these runs for you.

​
Your cloud computer’s disk is low or full

Your squad works on its own cloud computer. When that computer’s disk runs low or fills up, every screen of the dashboard shows a banner. Squad checks the disk about every 30 minutes, and waits while an agent is in the middle of a reply. See Limits and defaults for when a disk counts as low or full.
  • Low: an amber banner, for example “Your cloud computer’s disk is 88% full (9.4 GB free). When it fills up, your agents stop working and new settings cannot be saved.” Hide for a day hides it for a day in this workspace, in this browser.
  • Full: a red banner that you cannot hide: “Your cloud computer’s disk is full” with the free space, and that agents can stop working and new settings and connections can fail to save until you free up space.
The owner also gets a bell notification from Squad when the disk becomes low and again when it becomes full. When it becomes full, the owner also gets an email with the subject “[Squad] Your cloud computer’s disk is full”. It has no switch of its own in Settings > Notifications, and it is not sent when all notification email is off for your workspace. Teammates see the banner. Each banner and the email also say: “Ask an agent to delete files it no longer needs, such as build caches and installed packages it can reinstall.” While the disk is full, a change to an AI account on AI Tokens, for example adding, replacing, or reordering accounts, or Disconnect, fails with “Your cloud computer’s disk is full, so this change could not be saved. Free up some space on it, then try again.” A failed Disconnect shows its message under the account’s row. Fix: select Ask an agent to free up space on the banner. The button shows only when there is an agent that you can use. It opens a new chat with your lead agent, or with another agent that you can use when you cannot use the lead, and a prepared message, which you can change before you send it. The message asks the agent to find what uses the most space, delete what can be rebuilt (build caches, packages it can reinstall, old downloads), and ask you before it deletes anything else. The banner goes away after Squad checks the disk again and finds enough free space.

​
Rate limits and fallback

A rate limit means that your AI provider refuses calls for a time, because your plan reached its limit.
  • On AI Tokens, an account that hit a limit recently shows Rate-limited · cooling down.
  • After 3 or more rate-limit failures in a row, Health, Home, and Missions show “[provider] is rate-limiting your agents.” For the exact rule, see Limits and defaults.
What happens next depends on your setup:
  • With more than one account for a provider, Squad tries the accounts in the order on the provider card, highest first. Claude Code is different: it uses one account at a time and moves to the next account when the active one reaches its usage limit. AI Tokens then shows the new account as IN USE. The run continues once on the new account. If it had already done work, the agent is told to continue from where it stopped and not to repeat what is done. See Limits and defaults.
  • With fallback models in the Model routing card on AI Tokens, a run that fails or hits a rate limit moves to the next model in the list.
  • With no fallback models, the card says “No fallbacks are set. When the default model fails, runs on it fail until it recovers.”
The banner says that runs that hit the limit are not run again. Check that their work finished. A failed chat reply needs your Retry. See Which runs retry. Fix:
  1. Wait for your provider’s limit to reset.
  2. To avoid the problem, add a second account or a fallback model. See Choose models and manage AI accounts.

​
Time limits

Squad does not stop a run because it takes a long time. On Agent Configuration > Workspace defaults, the Run time limit card says “Runs have no time limit.” and shows No limit. You cannot set a time limit. Each scheduled task that you create in the dashboard gets 24 hours, and the task editor has no time limit field. A run can still stop for one of these reasons:
  • It showed no progress for 15 minutes. Squad stops a run that sends no progress signal, such as a tool call, for that period. Squad does not stop a run while a tool call or a memory compaction is in progress. Runs on a Claude Code account are not stopped this way. See Limits and defaults.
  • A time limit from an earlier setting is still saved. Runs that take longer than that limit are stopped, even while they are still working. If the workspace has no saved value, the card shows 10 minutes, and chat runs stop after 10 minutes.
  • A scheduled task that an agent created can have no time limit saved. Its runs then stop after 60 minutes, until the workspace computer restarts and raises the limit to 24 hours.
To remove a time limit that is left from an earlier setting:
1

Open the timed-out run

Open the run detail with the label Timed out.
2

Open the limit

Select Check the time limit →. For a scheduled run, the editor of that task opens in Agent Configuration. For other runs, the Workspace defaults tab opens.
3

Scheduled task runs

The task editor says “This task stops after [time].” Select Raise to 24 hours, then select Save changes.
4

Other runs

The Run time limit card shows the limit and says “This limit comes from an earlier setting. Runs longer than it are stopped even while they are still working. Remove it, then save.” Select Remove the limit. Then select Save changes in the bar at the bottom.
A save on Workspace defaults restarts the workspace computer, which interrupts runs in progress. The workspace computer also raises every scheduled task limit below 24 hours to 24 hours each time it restarts. For more about scheduled tasks, see Schedule recurring work.

​
Interrupted runs

A restart of the workspace computer ends every run in progress. Settings changes and routine maintenance cause restarts. Routine maintenance waits while an agent is working, for a limited time. A save on a settings screen restarts the computer at once. The run detail shows Interrupted. Unless Squad continued the run as described below, it says “Routine maintenance interrupted this run before it finished. It is not run again, so check that its work got done.” When it can, Squad gives a chat message that a restart interrupted back to the agent on its own. The chat then shows “[name] was interrupted by a restart and is picking up where it left off.” The agent continues in the same conversation, sees what it already did, and is told not to repeat an action that already completed, such as a sent email. Squad does this a limited number of times for one message. See Limits and defaults. When the message reaches the agent again, the run that the restart cut off shows “A restart interrupted this run; Squad resumed it automatically.” on Runs, and its row on Health is tagged resumed. The continued work is a new run. You do not need to send the message again. Squad does not continue a message that you stopped, edited, or reset, or a message that it cannot continue safely. The chat then shows “The workspace runtime restarted while handling this message, so it may not have been processed. Send it again if nothing arrived.” Check the work, then send the message again if it is not done. For other runs, Squad does not start the run again. Check the agent’s work. If it is not done, ask the agent to continue.
Ask the agent
A scheduled run that a temporary provider error stopped is different. Squad runs it once more, and its detail says “A model-provider hiccup interrupted this run; resumed automatically.” On Health, its row is tagged resumed.

​
Which runs retry automatically

Do not expect every failed run to run again. Squad retries only these cases: Squad does not retry these cases:
  • An expired or invalid AI connection
  • A usage limit or quota error from your provider, except a Claude Code usage limit that switches to your next account (see above)
  • A rate limit when you set no fallback models. A rate limit never switches Claude Code accounts.
  • A run where all models in the fallback list failed
  • A chat reply that failed. Select Retry yourself.
  • A manual run or a chat run that failed with a provider error
  • A run other than a chat message that a restart interrupted

​
A mission shows “Stuck for … · no progress”

Symptom: a mission card on Missions shows “Stuck for” a time, then “no progress”. Cause: the mission had no update for a long time. The agent can be waiting for your answer, or its runs can be failing. Fix:
  1. Open My Tickets and answer any open ticket for that mission. See Answer tickets in My Tickets.
  2. Check Health for failed runs of the agent.
  3. Open the mission and ask the agent for a status update in the mission thread.

​
An agent shows “Paused”

Symptom: the agent’s profile shows “Paused by” and a name, or “Paused”, at the top. Its scheduled tasks do not run, and a chat message to it fails with “[name] is paused. Resume [name] to send a message.” Cause: you, an Administrator, your lead, or the agent itself paused the agent. The profile names who paused it (right after you pause it yourself, Paused by you). A pause from before September 30, 2026 shows only “Paused”. A paused agent does not start new work from the dashboard. Its scheduled tasks are off, the dashboard does not send it chat messages, and group chats do not wake it. @mentions, mission assignments, and emails for the agent wait until it is resumed. Messages from Telegram, Discord, or Slack still reach it. Fix: select Resume, or ask your lead to resume the agent. The agent then gets the @mentions, mission assignments, and emails that waited. For more, see Add, change, and pause agents. If the profile shows “Needs attention” and a reason, read the reason and follow the fix on this page.

​
A new agent does not answer yet

Symptom: your lead just added an agent. A group chat refuses it with “[name] is still being set up and cannot join a group chat yet.” A Telegram, Discord, or Slack bot refuses it with “That agent can’t answer chats yet — pick another.” Cause: a new agent needs about 2 minutes before it can work. See Limits and defaults. Fix: wait a few minutes, then try again. Until then, work that you send to the new agent can go to the lead instead.

​
An agent works on the wrong thing

Symptom: an agent started work that you did not want, or a run takes a wrong turn. Cause: the agent understood the message differently than you meant, or it chose a path that you did not expect. A run goes on until it finishes unless someone stops it. Fix: stop the run. In chat, select Stop under the agent’s working message. In Runs, select Stop on the run’s row. Then select Stop the run. You must be the owner or an Administrator. The run shows Stopped, and Squad does not count it as a failure. Then tell the agent what you want instead. See Stop a run. If Stop says “The stop didn’t go through, so the run is still going. Press Stop to try again.”, select Stop again. A run on a Claude Code account cannot be stopped and finishes on its own. If Stop says “This squad’s computer needs an update before Stop works. It updates on its own, so try again later.”, wait and try again later.

​
Ask your lead to check

Your lead agent can look at the problem from inside your workspace.
Ask your lead

​
Still stuck

Email support@squad.so. Give the agent’s name, the time of the run from Runs, and the text from Technical details. See Get help.