ORCHESS — multi-agent development flow with declared roles

Context

ORCHESS is a multi-agent orchestration system under active development.

During its construction, the operator (Hernán Alfredo Capucci) works daily with multiple AI agents in a structured workflow: one agent builds, another evaluates.

This document records the manifests used in the author’s own workflow.

Agent 1 — ORCHESS Builder (Claude Code)

Role: construction and implementation under operator direction.

Autonomy: Level 2 (scoped — bounded workflows within declared constraints).

Risk posture: moderate.

Full manifest: claude-orchess-builder.json

Agent 2 — ORCHESS Evaluator (ChatGPT)

Role: architectural review and critical analysis of outputs produced by other agents.

Autonomy: Level 1 (supervised — operates only under explicit operator instruction).

Risk posture: low.

Full manifest: gpt-orchess-evaluator.json

Declared differences between the two agents

These agents operate with different autonomy levels, risk profiles, and explicitly distinct roles.

They do not share internal state or awareness of each other’s declared boundaries. The operator maintains that coordination layer.

Agent Manifest makes this structure explicit prior to interaction — not as runtime enforcement, but as a declarative layer.

What Agent Manifest Clarifies in This Workflow

Without declaration: agents produce outputs in isolation, compared only by result.

With declaration: agents operate with explicit roles, constraints, and stopping authority — allowing comparison at the level of intention, scope, and responsibility.

---

Case documented: March 2026
Specification DOI: https://doi.org/10.5281/zenodo.18833956 (Agent Manifest v1.0). This page has no DOI of its own.