Troubleshooting¶

Signing in creates your Agentic Unit user record, but it does not assign a team. Ask an Agentic Unit administrator to add you to the correct team and confirm that the team has at least one enabled agent.

If you cannot sign in, confirm that your account belongs to the aca-users or aca-admins Cognito group for the environment.

Agentic Unit Cannot Access A Repository¶

Check that:

The platform GitHub App is installed for the organisation.
The installation includes the repository.
The repository is linked to your Agentic Unit team.
The repository name and starting branch in the request are correct.
Branch protection is configured for important branches.

Agentic Unit runs in an isolated remote workspace. Giving it access to a directory on your laptop does not make that directory available to a hosted job. Use the CLI local workflow when you intentionally want Agentic Unit to work from a local checkout.

A Job Remains Queued¶

A QUEUED response means the platform accepted the job. It can wait while the team has reached its concurrency limit or while capacity is being allocated.

Keep the requestId, check the job in chat or Backoffice, and wait for its status to change. If it remains queued unusually long, send the request ID and environment to the Agentic Unit platform team.

Chat Lost Connection¶

Reconnect and open the job again. In the CLI, use the jobs command to find the job and resume it. Completed or failed work can be inspected with the result command. If the job is still running but no longer useful, interrupt or stop it instead of starting duplicate work.

A Pull Request Review Did Not Start¶

Confirm that:

The comment uses the mention configured for the environment, such as @ACA-DEV, @ACA-INT, @ACA-CONS, or @ACA-PROD.
It is a new comment on a pull request, not a regular issue or an edit to an old comment.
It was posted by a person rather than a bot.
The GitHub App installation includes the repository.
The repository is linked to your team.

See GitHub Pull Request Reviews for the full trigger rules.

An Automation Returns 401 Or 403¶

401 usually means the bearer token is missing, malformed, revoked, or sent to the wrong environment.
403 usually means the token is valid but is not allowed to use the requested team or resource.

Confirm that the URL, payload, and Bearer credential all belong to the same environment. Development Jira uses its legacy webhook key with /webhook/jobs/jira-events; Production Jira uses a scoped trigger token with /webhook/jobs/v2/jobs/dispatch. These credentials are not interchangeable.

Credentials are shown only when created. If the secret is unavailable, revoke it and create a replacement instead of trying to recover it from logs.

A Command Or Test Failed¶

Read the command output in the job. A repository may require dependencies, environment variables, build tools, or network resources that are not available to the runtime. Give Agentic Unit the repository's setup and verification instructions, or update the repository documentation so future jobs can discover them.

Never paste credentials into a prompt. Configure required secrets through the approved Agentic Unit secret integration.

Agentic Unit Made An Unexpected Change¶

Do not merge the result. Review the branch or local patch, keep only the hunks you understand, and give Agentic Unit a narrower follow-up with explicit constraints. Branch protection and human review remain the final safety boundary.

Getting More Help¶

Use your existing Agentic Unit platform-team contact route. The dedicated Service Desk request type is not yet live.

Include:

Environment and team
requestId or job ID
Repository, pull request, or work item link
Approximate time and observed status
Relevant error text with secrets removed

Do not send trigger tokens, passwords, access keys, or other credentials.