Reusable, grounded prompt templates per role and task.
Templates turn good prompts into repeatable workflows. IOModel ships starter templates for the three roles that work in a spec — Analyst, Architect, and Developer — each already grounded in the spec’s live context and the IOModel schema. You point AI at a file, section, or model object, not an empty chat window.
Templates by role
| Role | Template | Purpose |
|---|---|---|
| Analyst | Extend a requirement | Add goals, scope, or NFRs to an ADR/SRS section |
| Architect | Review against the model | Check a spec against its objects and list conflicts and missing links |
| Developer | Implement from the spec | Generate a skeleton from the embedded API/ERD contract |
Example: Analyst — extend a requirement
Extend ADR voice-checkout § Goal:
- Add an NFR for streaming latency and the failure modes to handle.
- Keep it consistent with the embedded workflow and ERD.
- Update the section in place; do not restructure the document.Example: Architect — review against the model
Review specs/adr-voice-checkout.mdx against model product.speech_service:
- List conflicts between the spec and the current model.
- Flag missing links and orphan objects near speech_service.
- Produce the findings as a spec section, not free-form prose.Example: Developer — implement from the spec
Implement § API from adr-voice-checkout.mdx:
- Generate a service skeleton from the embedded gRPC/OpenAPI contract.
- Keep the Shopper → Mobile App workflow and ERD constraints intact.
- Reference the model object paths the spec links to.Using templates well
Start from a template
Pick the template that matches the role and task.
Fill in specifics
Add the system description, scope, and constraints.
Keep the guardrails
Retain the schema reference, validation step, and dry-run for mutations.
Save what works
Capture effective templates so the team reuses them.
Mutations stay safe
Templates that change the model should request a dry-run plan first. Review the plan, then apply.
Build a small library of team templates. Shared, grounded prompts make AI results consistent across people and projects.
Last updated on