OpenSkillIndex← back to search
claude-code

LLMs-Universal-Life-Science-and-Clinical-Skills- openai-codex-agents

Build and operate OpenAI-first coding and agent workflows using the Responses API, current GPT and Codex models, Agents SDK, MCP/connectors, deep research, and approval-aware tool execution. Use when you need long-horizon software agents or OpenAI-native multi-agent orchestration.

install
source · Clone the upstream repo
git clone https://github.com/mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills-
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills- "$T" && mkdir -p ~/.claude/skills && cp -r "$T/Skills/Agentic_AI/OpenAI_Codex_Agents" ~/.claude/skills/mdbabumiamssm-llms-universal-life-science-and-clinical-skills-openai-codex-agent && rm -rf "$T"
manifest: Skills/Agentic_AI/OpenAI_Codex_Agents/SKILL.md
source content

OpenAI Codex Agents

Use this skill when the job is best served by OpenAI-native agent building rather than a generic framework wrapper.

Workflow

  1. Start with the Responses API unless you have a hard compatibility reason not to.
  2. Pick the model deliberately: 
    gpt-5.4
    for highest-quality reasoning, 
    gpt-5.4-mini
    for fast subagents and high-volume loops, current Codex-class models for long-horizon coding, deep-research models for multi-source synthesis, and 
    computer-use-preview
    only when GUI control is essential.
  3. Use the Agents SDK when you need handoffs, guardrails, sessions, tracing, or multi-agent composition.
  4. Use MCP/connectors only with explicit approval policy, domain trust, and auditability in mind.
  5. Validate the workflow on a real repo or real task with a small eval set before wider rollout.

Guardrails

  • Default to approval for sensitive MCP or connector actions.
  • Record the exact model, reasoning setting, and tool policy used in tests.
  • Keep coding agents on a bounded write scope and enforce diff review before merge.
  • Prefer official OpenAI SDKs and docs over third-party wrappers when behavior matters.

Output Requirements

  • State the chosen model and why.
  • State whether the workflow uses Responses only, Agents SDK, or a specialized research/computer-use surface.
  • State the tool/approval policy and one rollback trigger.