AI Agent System Prompt Designer skill for Claude and Codex
A production-grade agent system prompt with role, tool rules, edge-case policy, guardrails, and an output contract, plus three test inputs aimed at its…
Install it once and Claude or Codex runs it for you. When your request matches, your AI loads the skill, asks for anything it is missing, and works through the template itself. Prefer to fill it in by hand? See the full prompt template.
Download the skill (.zip) · View SKILL.md
When it is used
Use when an engineer is building a tool-calling agent and needs a system prompt covering tool rules, edge-case policy, and a strict output contract. Do not use for a simple chatbot persona or tuning a prompt for a single one-off task. Related topics: AI, Technical, Coding.
What it will ask you for
- Agent name and one-line job (free text, e.g. "Name and job")
- Tasks it must handle well (longer free text)
- Target model type (choose one: Chat model, Tool-calling model, Small local model, Model-agnostic)
- Tools it can call (longer free text)
- Behaviour when uncertain (choose one: Ask the user, Pick the safest option, Refuse and explain)
- Things it must never do (longer free text)
- Guardrails to include (choose any: Cite sources, Stay in persona, Admit uncertainty, Keep responses short, Escalate to a human when stuck)
- Strict output format such as JSON or a template (choose one: Yes, No)
How it works
- State the role and objective in a few sentences, and use the tricky example to define the hardest case.
- Write tool usage rules: when to call each tool, when not to, and the action on failure.
- Write the decision policy for uncertainty and edge cases, matching the chosen uncertain behaviour.
- Add guardrails each with a short reason, and the output contract with an example if strict format is on.
- Write three test inputs aimed at the likeliest failures, each with a clear pass condition.
What you get
The complete system prompt in sections (role, tools, decision policy, guardrails, output contract), followed by three test inputs each with a pass condition.
What it watches out for
- Describing tools by name only, so the agent calls them in the wrong situations.
- Writing guardrails as a list of prohibitions with no policy for the cases between them.
- Skipping the failure path for tools, which leaves the agent looping or inventing results.
How to install
Claude.ai and the Claude desktop app
- Download the skill .zip.
- Open Settings, then Capabilities, and find Skills.
- Choose Upload skill and select the .zip.
- Ask Claude for help with the task. It loads the skill when your request matches.
Claude Code
- Download the skill .zip.
- Unzip it into ~/.claude/skills/ (or .claude/skills/ inside a project).
- Start a new session. Claude Code loads the skill when your request matches.
Codex
- Download the skill .zip.
- Unzip it into ~/.agents/skills/ (or .agents/skills/ inside a project).
- Codex picks it up on its own. Type $ and the skill name to use it, or just ask for the task.
Cursor
- Download the Cursor rule (.mdc).
- Save it in .cursor/rules/ inside your project.
- Cursor's agent applies the rule when your request matches its description.
AGENTS.md (Codex, Cursor, and other coding agents)
- Copy the AGENTS.md section.
- Paste it into AGENTS.md at the root of your repository, creating the file if needed.
- Coding agents that read AGENTS.md follow it when your request fits.