Skip to content

Author agent prompts

Command: /coding-agents-prompting-flow · ← All scenarios · User guide

Author or adapt prompts for AI coding agents — skills, subagents, workflows, rules — through a discover → brief → blueprint → draft/harden → simulate → validate pipeline.

Use this when you’re creating, refactoring, reviewing, or porting a prompt: a new skill, an agent, a workflow, a rule, or migrating a prompt from one IDE/agent to another.

Not for: writing application code (Write or change code).

Running it

/coding-agents-prompting-flow Author a new skill for API contract testing
/coding-agents-prompting-flow Refactor the debugging skill for brevity and stronger gates
/coding-agents-prompting-flow Adapt this Claude prompt for Cursor

Like requirements authoring, this expects a strong model (Opus / GPT-5.6-sol-class or similar). The agent will ask you to switch if yours is too small.

How it works

It treats prompts like software. It extracts a Prompt Brief you approve, designs a blueprint, then drafts and hardens each prompt one at a time, simulates realistic runs, and validates that the result traces back to your intent.

flowchart TB
    Start(["/coding-agents-prompting-flow + request"]) --> Disc["Discover context & references, <br>extract intent"]
    Disc --> G1{"You approve <br>the Prompt Brief"}
    G1 -->|revise| Disc
    G1 -->|approve| Blue["Blueprint the structure"]
    Blue --> Draft["Draft a prompt"]
    Draft --> Harden["Harden & edit"]
    Harden --> Pass{"Passes <br>review?"}
    Pass -->|no| Harden
    Pass -->|more prompts| Draft
    Pass -->|all done| Sim["Simulate realistic runs"]
    Sim --> Risk{"Major <br>risk?"}
    Risk -->|yes| Blue
    Risk -->|no| Val["Validate the prompt set"]
    Val --> G2{"You give <br>final approval"}
    G2 -->|request changes| Draft
    G2 -->|approve| Done(["Validated prompt set"])

    classDef step fill:#dae8ff,stroke:#3674b5,color:#102a43;
    classDef gate fill:#fff2cc,stroke:#b58b00,color:#493800;
    classDef done fill:#e5f7e8,stroke:#35834a,color:#153b20;
    class Disc,Blue,Draft,Harden,Sim,Val step;
    class G1,Pass,Risk,G2 gate;
    class Start,Done done;

The draft/harden/edit loop is an automated review pass (not something you sit through). Adaptation requests — porting between agents or rule formats — load extra guidance for that.

What you’ll be asked to do

Provide the request plus any existing prompt, constraints, and audience; resolve the open questions; approve the Prompt Brief before design; and give final approval before anything is saved.

What it creates

Analysis artifacts in the plan folder (prompt-brief.md, open-questions.md, blueprint.md, and a validation pack), and the finished prompts written to their target folders. Small jobs may just return the result in chat. A state file tracks progress.

Get help to understand Rosetta’s own skills and agents · contributing prompt changes is covered in CONTRIBUTING.

Sources