Coding Agents Prompting Flow

TL;DR

Use Coding Agents Prompting Flow to author or adapt prompts for AI coding agents under explicit phase contracts and approval gates. The workflow stays thin on purpose: it discovers context, approves a Prompt Brief, designs a Blueprint, runs draft and hardening loops, simulates realistic usage, then validates the final prompt set against the original intent. It produces state plus prompt-authoring artifacts such as Discovery Notes, Reference Set, Prompt Brief, Blueprint, Draft Prompt Set, Prompt Set, Simulation Notes, and a Validation Pack. The hard approvals are the Prompt Brief gate and the final approval before persistence.

When To Use This Workflow

• Create prompts, rules, or instruction sets for coding agents.
• Adapt an existing prompt from one agent, IDE, or platform to another.
• Standardize a prompt family before storing it in a repo or prompt catalog.
• Validate that a prompt set still matches the original operating intent after edits.

When Not To Use This Workflow

• Do not use it for implementation work. Use Coding Flow.
• Do not use it for product or system requirements authoring. Use Requirements Documentation Authoring Flow.
• Do not use it for general Rosetta capability questions. Use Self Help Workflow or Usage Guide.
• Do not use it to onboard an external codebase for later reuse. Use External Library Flow.

Before You Start

• Provide the target outcome: create new prompts, adapt an existing prompt, or validate a prompt family.
• Provide the source prompt when one already exists.
• Name the target agent, IDE, or framework the prompt set must serve.
• Point the agent at any required prompt-family references, operating rules, schemas, or examples.
• Plan to review the Prompt Brief, blueprint tradeoffs, simulation findings, and final validation pack carefully.
• For shared Rosetta setup and global customization, keep Usage Guide current. This page only covers workflow-specific preparation.

How To Start

Create a coding workflow prompt for our internal AI agent.

Adapt this Claude coding-agent prompt for Cursor and preserve the approval model.

Write prompts for our onboarding automation agent and include validation artifacts before persistence.

Review this prompt family, harden it, simulate realistic runs, and return a final prompt set plus validation pack.

How Rosetta Shapes This Workflow

Rosetta provides instructions. Coding agents act on them. Rosetta itself does not see your prompt files, source code, or project data.

For this workflow, the always-active Rosetta behavior changes the UX in visible ways:

• The workflow stays sequential. It does not skip from a rough idea straight to final prompts.
• Approval is tied to artifacts, not vague conversation. The Prompt Brief must be approved before blueprinting, and final approval is required before persistence.
• Prompt authoring and prompt review stay inside the workflow contract. Drafting, hardening, editing, simulation, and validation are separate phases with separate artifacts.
• Reference loading is phase-scoped. The workflow explicitly tells the agent to load only what the current phase needs.
• State is persistent. Each phase records status and evidence in coding-agents-prompting-flow-state.md.

Workflow At A Glance

Workflow Overview

flowchart TD
    A["Start prompt authoring request"] --> B["Phase 1<br/>Discover"]
    B --> C["Discovery Notes + Reference Set"]
    C --> D["Phase 2<br/>Extract and Intake"]
    D --> E{"Prompt Brief approved?"}
    E -- "No" --> D
    E -- "Yes" --> F["Phase 3<br/>Blueprint"]
    F --> G{"Blueprint ambiguity or tradeoff stop?"}
    G -- "Yes" --> F
    G -- "No" --> H["Phase 4<br/>Draft Loop"]
    H --> I["Phase 5<br/>Hardening and Edit Loop"]
    I --> J{"Loop stalled or conflicts unresolved?"}
    J -- "Yes" --> I
    J -- "No" --> K["Phase 6<br/>Simulate"]
    K --> L{"Major behavior risk?"}
    L -- "Yes" --> I
    L -- "No" --> M["Phase 7<br/>Validate"]
    M --> N{"Final approval before persistence?"}
    N -- "No" --> I
    N -- "Yes" --> O["Persist final prompt set"]

    style A fill:#dbeafe,stroke:#1d4ed8,color:#0f172a
    style B fill:#dcfce7,stroke:#16a34a,color:#14532d
    style C fill:#dcfce7,stroke:#16a34a,color:#14532d
    style D fill:#dbeafe,stroke:#1d4ed8,color:#0f172a
    style E fill:#fef3c7,stroke:#d97706,color:#78350f
    style F fill:#dbeafe,stroke:#1d4ed8,color:#0f172a
    style G fill:#fef3c7,stroke:#d97706,color:#78350f
    style H fill:#dcfce7,stroke:#16a34a,color:#14532d
    style I fill:#dcfce7,stroke:#16a34a,color:#14532d
    style J fill:#fef3c7,stroke:#d97706,color:#78350f
    style K fill:#dbeafe,stroke:#1d4ed8,color:#0f172a
    style L fill:#fef3c7,stroke:#d97706,color:#78350f
    style M fill:#dbeafe,stroke:#1d4ed8,color:#0f172a
    style N fill:#fef3c7,stroke:#d97706,color:#78350f
    style O fill:#dcfce7,stroke:#16a34a,color:#14532d

Interaction Flow

sequenceDiagram
    actor U as User
    participant R as Rosetta Instructions
    participant O as Coding Agent / Orchestrator
    participant D as Discoverer
    participant P as Prompt Engineer
    participant A as Workflow Artifacts

    U->>O: Request prompt authoring or adaptation
    R->>O: Load workflow and phase contracts
    O->>D: Phase 1 discover
    D->>A: Write Discovery Notes + Reference Set
    O->>P: Phase 2 extract and intake
    P->>A: Write Prompt Brief + Open Questions
    O->>U: Request explicit Prompt Brief approval
    U-->>O: Approve or request changes
    O->>P: Phase 3 blueprint
    P->>A: Write Blueprint
    O->>P: Phase 4 draft loop
    P->>A: Write Draft Prompt Set
    O->>P: Phase 5 hardening and edit loop
    P->>A: Write Prompt Set
    O->>P: Phase 6 simulate
    P->>A: Write Simulation Notes
    O->>P: Phase 7 validate
    P->>A: Write Final Prompt Set + Validation Pack
    O->>U: Request final approval before persistence
    U-->>O: Approve persistence or request changes

Phases

Phase 1. Discover

Goal

Collect prompt-family context and only the references the rest of the workflow will actually need.

What you provide

• Request intent
• Optional source prompt
• Local prompt artifacts, schemas, or standards when they already exist

What the agent does

• Uses the discoverer subagent
• Finds project-local prompt-family artifacts and required references
• Stops for HITL if discovered context conflicts with intent or critical references are missing
• Updates coding-agents-prompting-flow-state.md

Artifacts

• Discovery Notes
• Reference Set

Phase 2. Extract and Intake

Goal

Turn the request into an explicit operating contract for the rest of the workflow.

What you provide

• Clarifications on goals, constraints, target platform, and success criteria

What the agent does

• Uses the prompt-engineer subagent
• Extracts requirements from the source prompt when present
• Produces Prompt Brief and Open Questions
• Waits for explicit approval of Prompt Brief
• Updates coding-agents-prompting-flow-state.md

Artifacts

• Prompt Brief
• Open Questions

Phase 3. Blueprint

Goal

Define the structure, actors, contracts, and boundaries of the target prompt set before drafting begins.

What you provide

• Approved Prompt Brief
• Feedback on tradeoffs if the blueprint exposes ambiguity

What the agent does

• Designs the blueprint from the approved brief
• Uses HITL if architecture or tradeoffs remain ambiguous
• Updates coding-agents-prompting-flow-state.md

Artifacts

• Blueprint

Phase 4. Draft Loop

Goal

Produce the first draft of each target prompt from the approved brief and blueprint.

What you provide

• Approved brief and blueprint

What the agent does

• Drafts the target prompts
• Keeps the loop phase-specific instead of mixing in validation claims early
• Records progress in coding-agents-prompting-flow-state.md

Artifacts

• Draft Prompt Set
• Optional change-log.md

Phase 5. Hardening and Edit Loop

Goal

Review and refine each draft prompt until it either passes or hits an explicit stop condition.

What you provide

• Clarifications only if the loop stalls or exposes conflicts

What the agent does

• Runs the hardening -> edit loop for each target prompt
• Keeps this as automated subagent review, not a substitute for human approval
• Stops for HITL if the loop stalls, conflicts appear, or intent becomes unclear
• Updates coding-agents-prompting-flow-state.md

Artifacts

• Prompt Set
• Optional change-log.md

Phase 6. Simulate

Goal

Test the prompt chain under realistic execution traces before final validation.

What you provide

• Any final scenario emphasis if certain runs matter more than others

What the agent does

• Simulates realistic runs
• Traces context and cognitive load across the prompt chain
• Stops for HITL when simulation exposes major behavioral risk
• Updates coding-agents-prompting-flow-state.md

Artifacts

• Simulation Notes

Phase 7. Validate

Goal

Prove that the final prompt set matches the original intent and carries explicit failure-mode and traceability evidence.

What you provide

• Final approval or requested changes before persistence

What the agent does

• Validates against intent, contracts, failure modes, and traceability
• Produces the Validation Pack
• Marks coding-agents-prompting-flow-state.md complete
• Requires final approval before persistence

Artifacts

• Final Prompt Set
• Validation Pack
• Optional validation-report.md

How To Review Results

Review this workflow like a prompt-systems owner, not like a casual editor.

• After Phase 1, check whether Discovery Notes and Reference Set include the real prompt-family artifacts and ignore irrelevant noise.
• After Phase 2, read the Prompt Brief line by line. If the brief is wrong, every later phase will optimize the wrong target.
• After Phase 3, verify the blueprint names the right actors, contracts, and scope boundaries for the target prompt set.
• After Phase 5, inspect whether the hardened prompt set still preserves the approved brief instead of drifting into a different operating model.
• After Phase 6, read Simulation Notes for context overload, missing handoffs, approval gaps, or brittle prompt chaining.
• After Phase 7, verify the Validation Pack contains checklist results, tests, failure modes, and traceability back to the original intent before you approve persistence.

Workflow-Specific Customization

• If the target prompt set belongs to a framework such as Rosetta, state the artifact families up front: skills, rules, workflows, commands, agents, and subagents. Discovery and blueprint quality improve when that inventory is explicit.
• If you are adapting from one agent or IDE to another, name the source platform, target platform, and the behavior that must stay unchanged.
• Provide existing prompt-family artifacts, schemas, examples, and review standards early instead of expecting the workflow to rediscover them from fragments.
• Keep the prompt-family reference set lean. The workflow explicitly says to load only references needed by the current phase.
• If the prompt set depends on project context files such as CONTEXT.md, ARCHITECTURE.md, or TECHSTACK.md, say exactly how those files are supposed to affect the prompt behavior.

Artifacts You Will Get

• coding-agents-prompting-flow-state.md in the feature temp folder
• Discovery Notes
• Reference Set
• Prompt Brief
• Open Questions
• Blueprint
• Draft Prompt Set
• Prompt Set
• Simulation Notes
• Final Prompt Set
• Validation Pack
• Optional change-log.md
• Optional validation-report.md

Common Mistakes

• Starting drafting before the Prompt Brief is approved.
• Treating Prompt Brief as optional after phase 2 even though every downstream phase depends on it.
• Collapsing draft, hardening, edit, simulation, and validation into one polished pass.
• Loading every reference file at once instead of keeping reference loading phase-scoped.
• Approving a prompt set before checking failure modes, traceability, and simulation findings.
• Adapting wording between tools without checking whether the approval model or role contracts changed.

Source Files

• coding-agents-prompting-flow.md
• coding-agents-prompt-authoring/SKILL.md
• pa-prompt-brief.md
• pa-validation-report.md
• pa-change-log.md