20. Error messages
[!ref] Full details in the User Manual, "Troubleshooting and FAQ".
Common messages and how to respond:
| Message | Meaning | Action |
|---|---|---|
| Service not ready | Subprocess still warming up | Wait 30 s; otherwise see error drawer |
| Tool call denied | Tool disabled / approval rejected / path out of bounds | Check tool policy and path scope |
| Model timeout | LLM hasn't responded for a while | Resend or switch to fallback Provider |
| Empty model response | LLM produced no output | Usually a filter trigger; rephrase the prompt |
| Context too long | Approaching model context limit | Trigger compaction or new session |
| Workspace locked | Already held by another process | Wait or restart app |
| Credential expired | Sign-in token invalid | Sign in again |
| Insufficient points (HTTP 402) | Account points exhausted | "Buy coupons" (coupon wording, consistent with account usage / server-side billing — not "top up / reset / balance"; no longer retried as transient) |
| Plan quota exhausted (HTTP 429) | Plan quota out | "Wait for refresh or upgrade" (no longer mistaken for rate limit) |
| Auth failed (HTTP 401 / 403) | Sign-in needed / key invalid | "Sign in again or replace your key" |
| No channel available (HTTP 503) | All channels unusable | Retry later or switch model |
| Upstream error (HTTP 502 / 529 / overloaded) | Upstream issue | Reports honestly; upstream auth failure no longer misleads you |
Provider endpoint not found (HTTP 404, endpoint_not_found_error) |
URL path wrong | Check the Provider Base URL |
| macOS App Translocation | App running in read-only temporary image (not dragged into Applications) | "Quit → drag to Applications → reopen → check for updates again" |
| Channel disconnected | IM channel offline | Auto-retry; rescan after 30 s if still failing |
| Binding code invalid | Wrong 6-digit code or expired (10 min) | Regenerate |
| Upgrade signature invalid | Upgrade package signature failed | Auto-rollback; try the in-app upgrade again |
| Skill signature invalid | Skill signature failed | Refuse execution; check with the publisher |
| Background task cap | 64 in-flight reached | Wait for GC or stop some manually |
| Selected model unavailable | The model picked for the session is not actually usable | Handled per Routing → Auto-repair: Off = report the error; Ask before switching = show the "Switch to a backup model?" bar (where no bar can be shown, report the error directly with the reason); Switch automatically = switch to another provider with the same-named model and say so. The whole session is never silently downgraded to a different model |
| Couldn't determine which model to use for this turn | Resolving the model before sending failed | Check that the provider and model are enabled, or pick again in the model selector |
| Service busy, continuing automatically in N seconds | Plan requests-per-minute limit / provider rate limit | Wait for the countdown, switch models or stop waiting; main turns, subagents and parallel tasks share rate-limit state |
| Service remains busy; restarting this turn automatically in N seconds (attempt n/3) | Repeated retries failed; waiting to restart the turn | Wait for recovery or stop; Retry is disabled during active turns and automatic retry; after restart the old error becomes a previous-failure-retried notice |
| 模型最近 N 次回复都是「…」…已自动停止 (the model's last N replies were all "…" … stopped automatically) | Repetition-spin halt (§19.10) | Add more specific instructions and continue, or try another model |
| ⚠️ 本次任务已执行 N 步（用户当前设定的上限） (this task has run N steps — the limit you currently set) | "Max steps" reached | Raise it or set 0 (unlimited), then "Continue" |
| Drive X is locked by BitLocker | The Windows drive holding the project is locked | Click "Locked by BitLocker" to bring up the system unlock dialog; it opens automatically once unlocked |
| Drive X is not accessible (not ready) | Drive not connected / not ready | Check it and retry |
| ❌ Compact session failed: … | The summary request ended in an error | History is kept as-is; retry /compact later |
| "…": self-repair gave up | An output-type heal couldn't correct it over several turns | Manual intervention: add instructions or switch models |
| Remote connection lost | SSH / remote workspace link dropped | The disconnect icon on the title and in the sidebar is clickable to reconnect; it spins while reconnecting and reports "Reconnect failed" on failure |
"Out of credit" errors carry direct purchase entry points: when the error type is
insufficient_balanceorquota_exceeded, the error block gains a "Buy vouchers" and an "Upgrade plan" button. No other error type gets them — throttling, auth failures and upstream outages are not fixed by spending money.