Error reference
Runtime errors, HTTP/API statuses, retries, usage limits, authentication, networking, request failures, and CLI conflicts.
This page diagnoses failures after Claude Code starts. For installation, PATH, login, and OAuth, see Installation troubleshooting. For unloaded settings, hooks, MCP, and memory, see Configuration debugging.
A Passion8 error may originate upstream, at the gateway, on your network, or in local configuration. Identify the category before retrying, changing models/settings, or contacting support.
#Quick diagnosis
| Error | Category | First action |
|---|---|---|
| API Error: 500 | Server failure | Retry later; check provider/gateway status |
| Repeated 529 Overloaded errors | Capacity | Wait or select another available model |
| Request timed out | Server or network | Split the task; adjust API_TIMEOUT_MS if appropriate |
| Server error mid-response | Interrupted stream | Inspect partial output, then continue |
| You've hit your session limit | Subscription/session limit | Wait for reset or use an available provider |
| Usage credits required for 1M context | Context eligibility | Disable 1M or meet official account requirements |
| Request rejected (429) | Rate limit | Wait, reduce concurrency/background agents |
| Credit balance is too low | Balance | Check Passion8 balance/key |
| Not logged in | Authentication | Official login or correct Passion8 variables |
| Invalid API key | Authentication | Check credential and variable |
| Unable to connect to API | Network | Proxy, DNS, firewall, Base URL |
| SSL certificate | TLS | Configure corporate CA or fix proxy certificate |
| Prompt is too long | Context size | Compact, remove large content, split tasks |
| Request too large | Payload size | Reduce attachments, images, PDFs, or tool output |
| selected model | Model availability | Select an available Passion8 model |
| thinking budget exceeds output limit | Thinking configuration | Lower thinking budget or increase output limit |
| --bg and --print conflict | CLI flags | Choose background or print, not both |
#Automatic retries
Claude Code retries temporary failures, including 5xx, 529, some 429s, timeouts, and disconnects. A final error often means retries were exhausted.
| Variable | Purpose | Recommendation |
|---|---|---|
| CLAUDE_CODE_MAX_RETRIES | Retry count | Lower for fast-failing CI |
| CLAUDE_CODE_RETRY_WATCHDOG | Extended capacity retries unattended | Use deliberately for overnight jobs |
| API_TIMEOUT_MS | Per-request timeout | Increase for slow proxies or long output when appropriate |
Certificate failures and errors after visible output are not blindly replayed. Preserving partial output avoids duplicate tool execution.
#Server errors
#500
A provider/gateway internal failure is not a malformed prompt or permission-rule problem.
- Wait one or two minutes and retry.
- Select another available model with /model.
- If isolated to Passion8, check its console and model supply status.
- If isolated to the official API, check Anthropic status.
#529
Capacity congestion is not the same as exhausted account credit. Avoid many background agents hitting one model simultaneously.
| Scenario | Recommendation |
|---|---|
| Interactive development | Retry later or use another available Sonnet/Opus model |
| Print-mode script | Bound retries and preserve input on failure |
| Batch/multiple agents | Reduce concurrency and queue work |
| One-hour-cache workload | Retry without unnecessary model/effort changes |
#Interrupted response
When partial output already exists, replaying the whole turn can repeat tool calls.
- Read the partial answer.
- Ask to continue from the last section.
- If a tool ran before interruption, inspect actual file/command state first.
#Usage and credit
| Error | Meaning | Passion8 action |
|---|---|---|
| Session/weekly limit | Official subscription limit | A gateway key does not alter subscription limits; gateway variables select the API route |
| 1M context credits required | Official eligibility missing | Use normal context and compact instead of forcing 1M |
| Temporary limiting requests | Provider throttling | Reduce concurrency and wait |
| 429 | Request or quota window limit | Reduce concurrent large-context requests |
| Credit balance too low | Key/account balance | Check balance, key state, and model permissions |
#Authentication errors
| Path | Variables/commands | Use |
|---|---|---|
| Official account | /login, OAuth, subscription | Official hosted surfaces and remote features |
| API/gateway | ANTHROPIC_BASE_URL, AUTH_TOKEN, API_KEY | Passion8, custom providers, CI |
Recommended Passion8 configuration:
export ANTHROPIC_BASE_URL="https://passion8.cc"
export ANTHROPIC_AUTH_TOKEN="sk-YOUR_PASSION8_API_KEY"
claude -p "Reply only with ok"Claude Code uses the root https://passion8.cc without /v1. The /v1 suffix is common in OpenAI-compatible clients, not this setup.
| Error | Check |
|---|---|
| Not logged in | For Passion8, ensure gateway variables reached the process; official subscription uses /login |
| Could not resolve authentication method | Avoid incomplete/conflicting OAuth, API-key, and gateway configuration |
| Invalid API key | Complete active key in the correct variable |
| Organization disabled | Official organization/policy; local settings cannot repair it |
| OAuth token expired | Reauthenticate the official account or use the intended gateway credentials |
| Cloud-provider credentials | Provider credential chain and profile |
#Network and certificates
| Symptom | Cause | Action |
|---|---|---|
| Unable to connect | DNS/proxy/firewall/URL | Check curl -I https://passion8.cc and proxy settings |
| TLS certificate | Corporate interception or missing CA | NODE_EXTRA_CA_CERTS or IT-managed trust |
| Cloud session host not allowed | Disallowed host | Use local execution or update the allowlist |
| Long wait before failure | Proxy handshake or packet loss | Investigate network; adjust timeout only as needed |
#Request errors
#Context too long
Prompt-too-long, request-too-large, and compaction failures often involve excessive context or attachments.
- Inspect /context.
- Summarize with /compact.
- Start a new session with /clear.
- Remove unnecessary logs, screenshots, PDFs, and MCP output.
- Split large repository work by directory.
#Models and thinking
| Error | Action |
|---|---|
| Selected model issue | Choose one actually available from the provider |
| Opus unavailable | Use Sonnet or an account/provider with Opus access |
| Organization restriction | Ask the administrator to change policy |
| Thinking unsupported | Use a compatible model or supported non-thinking configuration |
| Thinking budget exceeds output | Reduce MAX_THINKING_TOKENS or raise output budget on compatible models |
| Tool-use block mismatch | Inspect request structure and gateway body rewriting |
#Cache effects
Errors do not themselves change TTL, but recovery actions can alter cache identity.
| Action | Effect |
|---|---|
| Retry same model/effort | More likely to reuse the prefix |
| Switch model | Different cache identity; initial miss likely |
| Change effort/thinking | Can invalidate reuse |
| Continue after compact | Summary replaces history |
| Repair MCP/hooks | Schemas or instructions may change |
| Use clean configuration | Diagnostic only, not a normal hit-rate measurement |
See Commands and cache effects and Prompt caching.
#Official references
Support
Need help?
For setup, billing, or model issues, email us. Check the status page for uptime.
WeChat / QQ support is available at the bottom right.
