Troubleshooting
Start with the Condux page that reports the problem. Dashboard, Codebases, Integrations, and run pages preserve the relevant status and evidence so you do not need to reconstruct what happened from Slack or CI alone.
A codebase is not ready
Section titled “A codebase is not ready”Open the codebase from the dashboard and read the latest setup result.
- If a configuration pull request is open, review and merge it before expecting the codebase to accept work.
- If validation failed, inspect the reported preparation or readiness-check failure and correct the
.conduxconfiguration in the repository. - If a connection is unhealthy, reconnect it from Integrations.
Condux asked a question in Linear
Section titled “Condux asked a question in Linear”The run is waiting for a consequential product decision. Reply in the Linear conversation. You do not need to restart or delegate the issue again.
A run failed
Section titled “A run failed”Open the failed stage and begin with its summary and artifacts. Common causes include a repository test failure, preview deployment problem, unavailable external service, or a codebase environment that needs updating.
After correcting the underlying problem, delegate the work again. If the cause is unclear or cannot be resolved in your repository, contact the Condux team with the run link.
A provider’s usage limit is reached
Section titled “A provider’s usage limit is reached”Open Integrations. A connected AI provider whose subscription or credit quota is used up shows a Usage limit badge with the reset time, and its detail page lists each limit window with how much is used. Runs skip that provider and use your next connected provider instead; the run log and a Linear thought say which model took over. When no connected provider is available, the run fails before any work starts, and the Linear error names each provider and why. Anthropic and OpenRouter are also checked live at the start of each run, so a limit reached between checks is caught before a sandbox is created. When the provider reports no reset time, for example an OpenAI workspace that is out of credits, add credits and press Check connection to let runs try again.
To take a provider out of routing yourself, press Pause on its detail page. A paused provider is skipped the same way, keeps its credentials, and still answers Check connection; press Resume to put it back.
The connection itself is still valid, so there is nothing to reconnect. If another provider is connected, delegate the work again and Condux runs it there.
Mason does not respond
Section titled “Mason does not respond”Check the Slack page under Integrations to confirm that Slack is connected and Mason is installed. Start a conversation by mentioning the installed @mason user group in a channel. Replies continue in Mason’s thread without another mention.
If the integration says Slack needs to be reconnected, complete that step so the app has its current permissions.
A workspace or codebase secret is missing
Section titled “A workspace or codebase secret is missing”Workspace admins manage values shared by all codebases under Settings → Workspace secrets. Codebase-specific values live under that codebase in Codebases → Secrets. Members can inspect secret metadata and recent audit activity, but only admins can add, replace, or delete values. Stored values are write-only and cannot be recovered from the page.
A Slack run link does not expand
Section titled “A Slack run link does not expand”Confirm that the Slack workspace is connected to the same Condux workspace that owns the run. Private or cross-workspace run information is not exposed through link previews.
Still blocked
Section titled “Still blocked”Contact the Condux team at hello@condux.build with the workspace, codebase, and run link when available. Do not send model-provider credentials or other secrets.