Skip to content

Troubleshooting

hive ctl failures carry structured codes (full table in CLI): 2 usage, 3 app not running, 4 timeout, 5 not found, 6 unauthorized, 7 refused/unsupported.

  • no .hivemind/ found — run hive init --prefix XX in the repo (or a parent of your cwd owns one; find .. -name .hivemind to check).
  • .hivemind/ already exists — init refuses to clobber; use the tracker, or bare hive init to refresh the agentic stack.
  • Prefix rejected — must match ^[A-Z][A-Z0-9]{1,9}$ (e.g. MYP).
  • move refused — the issue still has sub-issues; re-parent them (hive link <child> --parent none) or use --copy.
  • Cross-repo id unresolved — register the other workspace (hive workspace register) and confirm with hive workspace list.
  • Exit 3 on canvas verbs — the app is not running: hivemind .. Issue verbs work without it; spawn/read/list do not.
  • Agent not spawnable — run hive agent detect in the workspace; install the CLI if missing. Distinguish spawnable runtimes from recognised-for-status ones (Agents).
  • Terminal empty after reboot — runtimes without resume (opencode) start fresh; others resume per their provider. Screen content replays from disk after a daemon restart either way.
  • Remote terminals ended — remote PTYs run in the app process; an SSH drop ends them and there is no remote daemon to reattach. Reopen the tile; use a local worktree when you need sessions to outlive connections.
  • Browser page reloaded after a view switch — known <webview> limit; tabs and history survive, scroll/form state does not.
  • Agent cannot browse — enable agent browser control in Settings, open a Browser tile, retry; the hive-browser skill reports what is missing.
  • View missing or disabled — check Settings → Plugins → Installed for runtime load errors and hive views list for package errors. A plugin that violated its budget stays disabled until the app restarts.
  • hive ctl read waits 100 s by default in ≤ 10 s slices; keep the total under your tool’s timeout. Exit 4 means the turn had not finished — call again to collect it.
  • hive ctl workflow --timeout is per worker (default 600 000 ms); give the calling tool a timeout longer than the slowest worker.
  • hive ctl open-review --timeout defaults to a 24 h ceiling; set an explicit value when calling from an agent.
Terminal window
hive upgrade # re-run the installer (no-op when current)
hive upgrade --dev # source install: git pull + rebuild

After upgrading, refresh workspace skills with hive add skill.

  • Dev-bridge needs tsx (node), not bun — bun drops @lydell/node-pty output on Linux.
  • Playwright/e2e: unset ELECTRON_RUN_AS_NODE, run under xvfb; the config isolates the profile.
  • HIVEMIND_SHELL_ENV=0 makes tiles inherit the app environment verbatim.

State is files: cat the issue, hive config get, hive views list. Fixes go through Contributing.