Skip to content

Troubleshooting and reporting blockers

  • No invitation found: reopen the original email link in the browser where you will finish signup. Barge removes the fragment from the URL and keeps it only in that page’s memory; a reload or copying the scrubbed URL loses it.
  • Rejected, expired, revoked, or already-used invitation: sign in if you already created an account. Otherwise request a fresh invitation through the reporting route below. A resent invitation supersedes the old link; use the newest email. Do not publish either link.
  • Address mismatch: use the invited email. Changing an alias is not a new grant.
  • Forgotten password: use reset your password on the sign-in page and the newest reset email. Never include the reset URL or password in a report.

Confirm that you authorized the intended GitHub account and installed the deployment’s App, and included the repository in its selection. Organization approval may be pending. You also need current write permission. Signing in alone does not install the App. Use Settings → GitHub to reconnect or manage installation access, then reopen New work.

If clone, push, or PR creation fails, confirm that the starting ref and PR base exist and read the permission or branch-policy error. A repository can be visible while a write operation is refused. Do not work around an organization policy by pasting personal or administrator credentials into a task.

Use the matching credential method for the selected agent. An API key is not a subscription login. Reconnect expired device/token connections, run the app’s connection test, and review your provider account’s credit, model access, and rate limits. A successful connection test does not guarantee that every model or later request is permitted. For throttling, wait before retrying. Never paste a provider’s raw error response into a public report; it may contain sensitive data.

Failed tasks, retries, and missing results

Section titled “Failed tasks, retries, and missing results”
  1. Open the session and read its latest stage, event, failure, heartbeat, and cleanup information. A queued session may be waiting for capacity; starting another copy does not resolve that. A terminal session should not keep consuming a worker.
  2. If you need to stop active work, use Cancel and wait for confirmation. Report failed cancellation or continued worker activity rather than repeatedly retrying.
  3. Fix the cause first: missing repository access, unavailable provider credit, invalid .barge.yml, failing setup, or an unreachable private service.
  4. For a failed session, use Retry when available. Retrying reuses its task and local-handoff artifact; it is not a fresh upload of edits you just made locally. If the local inputs must change, make a new deliberate handoff after inspecting the previous result. A retry can consume additional provider resources.
  5. Read the session’s PR link and review GitHub before starting again. A run can open a PR and then fail a later review or test. If there is no PR, keep the session ID for a private follow-up.

Failed local handoffs may retain their encrypted uploaded workspace for retry. See data and leaving before assuming cancellation deletes all history or that a retry uses new input.

The public Barge issue tracker is an available reporting route. It requires a GitHub account and is public. Do not post invitation/reset links, session links or IDs, email addresses, organization names, private repository names/content, credentials, device codes, or raw logs. Use it for a sanitized report or a request for a private follow-up channel before sharing account-specific diagnostics. No setup call or response time is promised.

Use this public-safe template:

Step: invitation / GitHub / provider / start / progress / cancel / PR review
Expected: a short description with no account or repository details
Observed: a summary of the failure, not a raw response
Time: approximate UTC time
Environment: browser/OS and CLI version if relevant
Reproduction: steps using a disposable example repository

Privately retain the session ID, failure stage, approximate time, CLI version, and the actions already tried. Share them only after a maintainer supplies a private route. Crop screenshots to generic UI states; never include the address bar, account menu, repository picker, credential field, code, or event output without reviewing every visible value. Do not upload your CLI config, .env, provider authentication files, network archives, or terminal history.