Troubleshooting

Troubleshooting

Start with shibaox doctor: it checks Node, git, Bun, the install, the daemon (and whether it is older than the CLI), the service, the keys, Telegram and the Claude login, and says what to do.

The daemon

• "No shibaox daemon is running" — any command starts it on demand; SHIBAOX_NO_AUTOSTART=1 prevents that. shibaox daemon start in the foreground shows why it fails to start; ~/.shibaox/daemon.log keeps the output of the service.
• "Daemon … is older than this CLI" — shibaox daemon stop; the service (or the next command) starts the new build.
• The service keeps restarting — tail ~/.shibaox/daemon.log. After a Node upgrade the launcher says the recorded node … is gone or is not the same ABI: run shibaox daemon install again. daemon status and doctor say when this is due.

Native modules and Node

• better-sqlite3 fails to build or load — a Node version other than the one it was built for. Installer checkout: shibaox upgrade (or cd ~/.shibaox/app && pnpm rebuild better-sqlite3). npm: npm i -g shibaox@latest. Without a prebuilt binary you need a toolchain (xcode-select --install) and npm i -g node-gyp.
• The launcher says the Node on the PATH has a different ABI — it still runs the CLI, but the daemon will fail on native modules: rebuild as above.

The dashboard

• "The dashboard needs Bun 1.3 or later" — install it from https://bun.sh; the text commands work meanwhile (shibaox runs, inbox, follow --json).
• "The dashboard needs an interactive terminal" — run it in a terminal, not through a pipe.
• "finding the org…" stays — the daemon did not answer for the default org; shibaox daemon status and daemon.log.

Models

• "model … is not offered by " — the provider's own listing does not have that name; /model in the dashboard shows what it lists.
• "cannot start: role … → provider … is not configured" — the role's tier points at a provider without a key: shibaox keys set <KEY>, or change the tier (shibaox tiers set strong …).
• A run costs $0 or has no context percentage — the model has no known price or window (a hosted provider without a listing). OpenRouter and local models are discovered; others use the catalog.
• Decide nodes always pick ship, judge and review checks fail — no callable decision model: set tiers.decision to a model ref another provider serves (for example openrouter/typesafe/jev-router) or TYPESAFE_API_KEY. Subscription-only orgs cannot call models from the daemon.

Runs

• "cannot use a worktree: project has no commits" — commit first, or --workspace inplace.

• paused_budget — shibaox resume <runId> --budget <higher>.

• The merge failed with "rebase … failed (conflicts)" — the base moved with conflicting changes; the branch is intact in its worktree, the base untouched. Resolve by hand or rerun.

• Tests or lint fail in a worktree with "command not found" — the tool is not installed there; the tests and lint checks treat that as a skipped pass, the merge node's tests too.## Telegram

• The bot does not answer — shibaox doctor runs getMe; the token must be in the vault (shibaox keys set SHIBAOX_TELEGRAM_TOKEN …) and chat_id must be your private chat. Groups are ignored.

• A message sent while the daemon was down was skipped — by design: only the last of the backlog is answered, and the chat is told.