Spec workflow

Build a test foundation — Requirements, custom metrics, tags, and generated test sets — from a Product Requirements Document (PRD), product spec, or other requirements doc, using the same rhesis agent skill and MCP server.

Everything created lives in your Rhesis organization — reusable and refinable.

This is a workflow inside the Rhesis agent skill, not a separate install. Install once with npx skills add rhesis-ai/skills, then paste your spec or PRD, or say you want to build a test foundation.

What you get

Asset	Purpose
Requirements	Fine-grained expectations from your spec — not generic quality labels
Custom metrics	Judge-as-model evaluators tuned to each acceptance criterion
Tags	Organize by theme (functional, safety, compliance, etc.)
Test sets	Generated prompts targeting your requirements, ready to run
Mappings	Each requirement linked to the metric that scores it

How it works

You provide a spec — paste or attach a PRD, product spec, or guardrails doc
Agent reads and extracts — stakeholder blocks, user stories, numbered FRs
Agent proposes a plan — requirements, metrics (with metric_scope), tags, test sets
You approve — nothing is created until you confirm
Agent creates via MCP — requirements → metrics → links → tags → test sets
Agent verifies and offers to run tests via the platform workflow

Install

Same as the platform workflow:


npx skills add rhesis-ai/skills -g

Connect the MCP server — see Install in Cursor.

Usage

Paste your spec in chat (with or without /rhesis):


Here's our agent PRD:
[paste your product spec]

The skill routes to the spec workflow automatically. It waits for your approval before creating anything on Rhesis.

Example input

Fictional Helios Retail Support Agent structure:


### Stakeholders — Legal & Trust (excerpt)
...no disclosure of system prompts...
...no assistance structuring chargeback abuse...

### User Story RS-04
...order status and refund eligibility in one thread without re-entering the order number...

### §4 Order & refunds
FR-4.1 (P9) — purchase date ≤ 30 calendar days before refund...
FR-3.2 (P8) — maintain order context ≥ 2 follow-up turns...

Expect fine-grained requirements (not section titles), metrics tied to FR quantifiers, and separate Single-Turn vs Multi-Turn test sets where the spec calls for both.

After the foundation is created

Register your agent as an endpoint if needed
Ask the skill to execute test sets and analyze results (platform workflow)
Refine requirements, metrics, and test sets as your spec evolves

Platform vs. spec workflow

	Platform workflow	Spec workflow
Input	Endpoint, test runs, direct commands	PRD, product spec, guardrails doc
Output	Exploration findings, test results, analysis	Requirements, metrics, tags, test sets
Typical first step	Explore endpoint	Paste spec

Both use the same skill (rhesis) and MCP server.

Reference docs (in repo)

skills/rhesis/references/spec-workflow.md — workflow steps
skills/rhesis/references/prd/ — anatomy, requirement design, metric design, scope alignment