Openai API Troubleshooting Skill for Codex

OpenAI API Troubleshooting

Use this for runtime OpenAI API failures after a request has already been made. Keep key setup in openai-platform-api-key and current guidance in openai-docs.

Routing

• Sandboxed or blocked outbound network access
• Treat DNS errors, connection timeouts, connection resets, or inability to reach api.openai.com before any API response as transport problems first.
• If the failure came from a sandboxed or restricted run, explicitly recommend retrying the actual request with network access enabled or outside that restriction; only classify API-side auth, quota, rate-limit, or model access after a concrete OpenAI response exists.

• Authentication or missing-key errors
• Route 401, invalid_api_key, missing OPENAI_API_KEY, or malformed-key cases to authentication.
• If a key must be created or configured, explicitly hand off to openai-platform-api-key; do not stop at generic "create a fresh key" advice.

• Quota or credit exhaustion
• Treat insufficient_quota, "current quota", "billing quota", "run out of credits", or "no balance left" as billing/quota exhaustion, not ordinary throttling.
• For exhausted credits, prompt the user to add credits and include Add API credits. Do not purchase credits or change billing settings.
• Link usage caps to https://platform.openai.com/settings/organization/limits.
• If ambiguous, say it may be credits or a spend limit and consult openai-docs; when useful, note that ChatGPT subscriptions and API billing are separate.

• Rate limits
• Route rate_limit_exceeded, requests-per-minute, tokens-per-minute, or retry-after guidance without quota language to throttling.
• Recommend pacing, batching, exponential backoff, or lower concurrency; do not suggest buying credits unless the error also indicates quota exhaustion.

• Model, project, or organization access
• Treat 403, model_not_found, org/project mismatch, or permission failures as reached-OpenAI-but-not-authorized cases.
• Check the model, project, organization, and key scope before guessing.

Rules

• Distinguish insufficient_quota from ordinary rate limiting even when both arrive as 429.
• Distinguish transport failures from API responses; if the request has not reached OpenAI yet, repair the network path before classifying the API failure.
• Prefer the concrete API error code and message over generic heuristics.
• Keep user-facing answers short: likely class, reason, next action.
• Do not rotate or create keys here.
• Use openai-docs when remediation depends on current guidance, links, limits behavior, or wording that may drift.