Codex skills for OpenAI API
OpenAI API skills cover the Responses API and what is built on it: Structured Outputs, function calling, the Agents SDK, Realtime voice, Batch, embeddings, fine-tuning, images and moderation. Codex picks one conversation-state strategy per thread, handles refusals and truncated responses before parsing, and migrates Assistants code, since that API shut down on 26 August 2026.
Start with these 3
- OpenAI Responses API Typed output items, deliberate state, clean streams Codex writes Responses API code that walks output items by type, keeps reasoning items when replaying history, and resends instructions on every turn. It picks one conversation-state strategy on purpose, rewrites Chat Completions shapes with a checklist, and treats truncated responses as failures instead of parsing half an answer.
- OpenAI Agents SDK Few agents, bounded runs, guardrails that block Codex designs Agents SDK workflows with the fewest agents that do the job, picks handoffs only when a specialist should own the conversation, and runs input guardrails before side-effecting tools can fire. Every run gets a turn limit, one state strategy and a named trace, so loops and leaks show up in tests instead of production.
- OpenAI Realtime Voice Voice agents that know when to stop talking Codex wires Realtime voice agents with the API key kept on the server, the right transport for each client, and turn detection chosen from how users actually speak. It handles barge-in by stopping playback and truncating unheard audio, closes the tool-call loop with response.create, and measures end-of-speech to first-audio latency instead of guessing.
10 more OpenAI API skills
-
OpenAI Structured Outputs
Typed JSON from the model, every time -
OpenAI Function Calling
Tool loops that finish and fail gracefully -
OpenAI Batch API
Half-price bulk inference that rejoins cleanly -
OpenAI Embeddings
One vector space, measured, never mixed -
OpenAI Image Generation
Generate, mask, iterate and store images -
OpenAI Fine-Tuning
Fine-tune GPT-4.1 only when evals prove it pays -
Assistants to Responses
Get off the shut-down Assistants API safely -
Prompt Caching and Cost
Lower the LLM bill without lowering quality -
Text to Speech
Natural narration with names pronounced right -
AI Content Moderation
Policy first, thresholds second, humans in the loop
General skills with OpenAI API guidance
Not built around OpenAI API, but they include steps or tools for it. Not part of the install-all command.
-
Codex GitHub Action
Codex in workflows without leaking the key -
Codex SDK Automation
Drive Codex from TypeScript with real guardrails -
Structured Extraction
JSON from messy text, with evidence and per-field scores -
Ollama Local LLMs
Local models with the context you think you have -
llama.cpp and GGUF
Pick the quant by perplexity, not folklore -
Speech to Text
Transcripts with speakers, timestamps and a WER -
Form Data Extraction
Invoice fields as validated JSON, not guesses
Every OpenAI API skill in one command
Installs these 13 into ~/.agents/skills/. Add --project after -- to install into the current repository instead.
The installer is a 104-line POSIX shell script. It fetches the skills you name from codexskills.dev and unpacks each into its own folder in your skills directory. No sudo, no telemetry, and every skill page shows the full SKILL.md first. Read install.sh
curl -fsSL codexskills.dev/install.sh | sh -s -- openai-responses-api openai-agents-sdk openai-realtime-voice openai-structured-outputs openai-function-calling openai-batch-api openai-embeddings openai-image-generation openai-fine-tuning assistants-to-responses-migration prompt-caching-cost text-to-speech ai-content-moderation OpenAI API FAQ
Is there a Codex skill for migrating from the Assistants API to the Responses API?
Assistants to Responses inventories every beta.assistants and threads call, moves instructions and tools into versioned code, and rewrites runs and polling as responses.create with a capped function_call loop. Because the Assistants API shut down on 26 August 2026, it backfills Conversations from your own stored message history after you confirm the plan, and says plainly if no such copy exists. A replay set checks that the new path behaves like the old one.
How do I get reliable JSON from OpenAI models?
OpenAI Structured Outputs defines the schema once in Pydantic or Zod, calls responses.parse with strict mode, and models optional fields as nullable so strict mode accepts the schema. It checks for refusals and an incomplete status before trusting output_parsed, then validates business rules in code because schema-valid data can still be wrong. OpenAI Function Calling applies strict schemas to tool arguments and returns each result under its exact call_id.
Can Codex run thousands of OpenAI requests through the Batch API?
OpenAI Batch API builds JSONL files with stable custom_ids tied to source rows and validates them offline with scripts/batch_jsonl.py check before spending money. It designs around the limits of 50,000 requests and 200 MB per input file and the 24h completion window, joins output and error files by custom_id rather than position, and writes failed or expired requests to a retry file. Batch jobs run at half the synchronous price.
How do I install all 13 OpenAI API skills at once?
Pass every name to the installer in one command: curl -fsSL https://codexskills.dev/install.sh | sh -s -- openai-responses-api openai-agents-sdk openai-realtime-voice openai-structured-outputs openai-function-calling openai-batch-api openai-embeddings openai-image-generation openai-fine-tuning assistants-to-responses-migration prompt-caching-cost text-to-speech ai-content-moderation. Add --project right after -- to put them in the current repository's .agents/skills/ so your team gets them too.
Which OpenAI API skill should I try first?
OpenAI Responses API is the closest fit for most OpenAI API work. Mention it with $openai-responses-api in a prompt to load it on purpose; after that, Codex picks it up on its own whenever a task matches.