Error listings
Developer running OpenAI API calls behind NAT gateway experiences infinite hangs on non-streaming calls (Responses API / completions.create()). Server-side succeeds but client blocks forever. Needs connection timeout or keepalive fix. Includes evidence for OpenAI API troubleshooting demand.
Non-streaming OpenAI API calls hang indefinitely behind NAT gateway — server generates response but client never receives it Developer using OpenAI Python SDK streaming encounters JSONDecodeError when server sends SSE events containing only meta-fields (retry, id, event) without a data field. Needs robust parsing. Includes evidence for OpenAI API troubleshooting demand.
JSONDecodeError: Expecting value: line 1 column 1 (char 0) — in openai/_streaming.py line 259 json.loads(self.data) Fix Claude Code web client not performing OAuth discovery for MCP servers, causing connection failures when connecting to external authenticated services Includes evidence for Claude Code troubleshooting demand.
claude.ai skips OAuth discovery entirely — POST /oauth/token endpoint missing Fix Claude Code losing GitHub push permissions during extended coding sessions, preventing commit workflow Includes evidence for Claude Code troubleshooting demand.
Claude Code cloud environment cannot push: git push permission denied Fix multi-turn chat failures when using DeepSeek V4 Pro through LiteLLM proxy where assistant messages lose reasoning content on sequential turns Includes evidence for LiteLLM troubleshooting demand.
Assistant message missing reasoning_content — LiteLLM strips it when serializing response from turn 1 Fix LiteLLM proxy leaking internal _litellm_* parameters through virtual key rate limiting boundaries, breaking per-key rate limit enforcement Includes evidence for LiteLLM troubleshooting demand.
Virtual keys with tpm/rpm limits leak _litellm_* params — rate limit enforcement broken at proxy layer Fix OpenAI API 429 insufficient_quota when running AI coding tools (Codex, Cursor) that depend on API key authentication Includes evidence for OpenAI API troubleshooting demand.
insufficient_quota — You are out of money Fix Vercel deployment build failures caused by missing or misconfigured NEXT_PUBLIC_ environment variables in Next.js projects Includes evidence for Vercel troubleshooting demand.
Missing NEXT_PUBLIC_SUPABASE_URL — Vercel build command fails at deploy time Fix 403 permission_error when logging into Claude Code CLI — keychain entry deletion fails and login returns permission denied Includes evidence for Anthropic API troubleshooting demand.
API Error: 403 {"type":"error","error":{"type":"permission_error","message":"Permission denied"}} Fix Claude Code MCP server OAuth flow where PKCE state is lost between authenticate and complete_authentication steps Includes evidence for Claude Code troubleshooting demand.
"No OAuth flow is in progress for {provider}. Call mcp__{provider}__authenticate first" Fix Anthropic API 529 overloaded_error not being automatically retried, causing sessions to halt instead of retrying with backoff Includes evidence for Anthropic API troubleshooting demand.
overloaded_error: An error occurred while reading the stream: read tcp: i/o timeout Get stable machine-readable error codes mapped to SDK exception classes for OpenAI Responses API background mode failures Includes evidence for OpenAI API troubleshooting demand.
SDK does not raise typed exception for background mode failed runs — returns 200 OK with free-form error.code string Fix Vercel prebuilt deployment failure where Turbopack hashed server external aliases become unresolved in production runtime Includes evidence for Vercel troubleshooting demand.
Failed to load external module require-in-the-middle-<hash>: Cannot find module 'require-in-the-middle-<hash>' Fix Anthropic SDK / Vertex AI integration incorrectly rejecting valid-sized requests with 413 payload-too-large errors Includes evidence for Anthropic API troubleshooting demand.
Vertex AI incorrectly rejecting ~2MB requests with 413 '32MB limit exceeded' error despite payload well under stated limits Fix TPM rate limiting that bypasses multi-replica deployments; each pod enforces its own TPM limit independently doubling/quadrupling effective throughput beyond intended cap Includes evidence for LiteLLM troubleshooting demand.
Deployment-level TPM enforcement is per-pod not cross-pod — effective limit becomes tpm_limit × N_replica allowing uncontrolled token consumption Fix phantom BudgetExceededError in LiteLLM proxy caused by Redis spend counter leaks; end users randomly blocked despite being within budget limits Includes evidence for LiteLLM troubleshooting demand.
BudgetExceededError (HTTP 429): "Budget has been exceeded! EndUser=<id> Current cost: <inflated>, Max budget: 50.0" — DB spend shows $0.001 while Redis counter reports $50+ after ~4 min of normal traffic Fix streaming fallback mechanism in LiteLLM router that drops mid-stream errors instead of gracefully switching to backup model Includes evidence for LiteLLM troubleshooting demand.
Router.aresponses streaming bypasses mid-stream fallback — MidStreamFallbackError not handled, breaking streaming responses when primary model fails mid-stream How to handle background mode Responses API errors deterministically in Python SDK; map opaque error codes to typed exceptions for retry logic Includes evidence for OpenAI API troubleshooting demand.
Background responses failures lack a stable code/name that maps to an exception class — status="failed" returns opaque error.code (e.g. "server_error") with no SDK exception mapping; HTTP poll returns 200 OK so no typed exception raised User blocked by undocumented per-minute rate limits in Claude Code despite low overall usage; needs workaround or per-minute limit documentation Includes evidence for Claude Code troubleshooting demand.
API Error: Rate limit reached (session usage 6%, weekly 29%, no per-minute docs) Developer expects SDK to auto-retry on 429 but agent crashes outright; needs robust retry/backoff for production reliability Includes evidence for Claude Code troubleshooting demand.
Agent SDK crashes on 429 rate limit — no exponential backoff or retry implemented