---
schema_version: "newruntime-agent-readable-v0.2"
type: "post"
stable_id: "post:hermes-prototype-to-specialist-agent"
slug: "hermes-prototype-to-specialist-agent"
title: "A Specialist Agent Should Be Promoted From Repeated Work"
description: "A Hermes practitioner workflow suggests building specialist agents by observing repeated real runs, correcting drift, and extracting the stable procedure only after it works."
retrieval_nugget: "A Hermes practitioner workflow suggests building specialist agents by observing repeated real runs, correcting drift, and extracting the stable procedure only after it works. A useful practitioner rule for Hermes is to avoid designing a specialist agent in the abstract."
status: "published"
published_at: "2026-07-24"
updated_at: "2026-07-24"
record_date: "2026-07-24"
date_kind: "published_at"
topics: ["hermes","specialist-agents","agent-evaluation"]
source_urls: ["https://x.com/shannholmberg/status/2056043898804539405","https://hermes-agent.nousresearch.com/docs/","https://github.com/NousResearch/hermes-agent"]
visuals: [{"id":"specialist-agent-promotion-loop","kind":"editorial-diagram","role":"hero","src":"https://newruntime.com/images/posts/specialist-agent-promotion-loop.webp","alt":"A five-station loop turns observed real work and preserved traces into a narrow specialist agent contract with an exception route back to the orchestrator.","caption":"Specialization is a promotion loop: observe work, preserve evidence, extract recurrence, define a contract, and retain escalation.","credit":"New Runtime synthesis from the practitioner workflow and Hermes documentation","source_url":"https://x.com/shannholmberg/status/2056043898804539405","generated_with":"gemini-3.1-flash-image","width":1600,"height":900,"legend":[{"label":"Observe","description":"Let the general agent encounter the real task before inventing a specialist role."},{"label":"Preserve","description":"Keep inputs, decisions, failures, corrections, and completion evidence from repeated runs."},{"label":"Extract","description":"Separate recurring behavior from one-off cases and compress it into a narrow procedure."},{"label":"Contract","description":"Package instructions, tools, checks, retries, and a machine-checkable output boundary."},{"label":"Promote","description":"Route stable work to the specialist while exceptions return to the orchestrator."}]}]
routes: {"html":"https://newruntime.com/posts/hermes-prototype-to-specialist-agent/","markdown":"https://newruntime.com/posts/hermes-prototype-to-specialist-agent.md","json":"https://newruntime.com/posts/hermes-prototype-to-specialist-agent.json"}
source_format: "markdown"
---

# A Specialist Agent Should Be Promoted From Repeated Work

## Retrieval answer

A Hermes practitioner workflow suggests building specialist agents by observing repeated real runs, correcting drift, and extracting the stable procedure only after it works. A useful practitioner rule for Hermes is to avoid designing a specialist agent in the abstract.

A useful practitioner rule for Hermes is to avoid designing a specialist agent in the abstract. Start inside the main agent or orchestrator, run the workflow repeatedly, correct drift and mistakes, and extract a specialist only after the behavior becomes stable.

The often-cited range of four to ten guided runs is a field heuristic, not an official Hermes guarantee. The valuable part is the sequence:

1. let the general agent encounter the real task;
2. preserve the inputs, decisions, failures, and corrections;
3. identify the parts that recur;
4. turn those parts into instructions, tools, checks, and a narrow output contract;
5. keep a route back to the orchestrator for exceptions.

## Why this is better than naming roles first

A role such as "research agent" or "review agent" says almost nothing about what good work looks like. Repeated runs expose the actual boundary: which sources are acceptable, which decisions need a human, which errors recur, and what evidence proves completion.

Specialization is therefore a compression step. The agent is promoted from observed work rather than invented from a title.

## New Runtime Read

This is directly applicable to the newsroom. The X bookmark processor, primary-source resolver, site materializer, R2 backup, and OpenClaw handoff should first be run as one observable workflow. Once the same repair pattern appears several times, that repair can become a specialist skill with fixtures and acceptance checks.

The promotion gate should not be "it worked once." It should be:

- the task boundary is stable;
- drift has named failure modes;
- the output is machine-checkable;
- retries are idempotent;
- escalation to the owner or orchestrator is explicit.

The specialist agent is not a smaller personality. It is a tested operational contract.