open-orchestrator-intent

name	open-orchestrator-intent
description	NL goals to NormalizedGoal JSON for orchestrator.
version	1.1.0
author	OpenPro
license	MIT
platforms	["linux","macos","windows"]
metadata	{"hermes":{"tags":["openorchestrator","intent","classifier"],"category":"open-ecosystem","related_skills":["open-orchestrator-plan","open-dev-workflow"]}}

Open Orchestrator Intent

Classifies free text into NormalizedGoal JSON — no planning here.

When to Use

• POST /v1/goals with natural language only
• intent_classifier profile invoked
• Pre-plan normalization when objective is unstructured

Prerequisites

• Orchestrator :3050 — schema GET /v1/schemas/openorchestrator.normalized-goal.v1
• Profile intent_classifier from openagents openos init-profiles
• INTENT_CLASSIFIER_AGENT_PROFILE=intent_classifier on orchestrator

Procedure

1. Read task_context.response_schema_url if present
2. Parse user sentence + optional brain context
3. Emit JSON only (no markdown fences)
4. Orchestrator validates with Zod + policy before plan step

Response fields

Field	Notes
intent	Slug e.g. ticket_fix, mission
goal_class	engineering | mission | qa | sales | security
objective	One imperative sentence
risk_level	1–4; ≥3 → approval_required
ticket_key	Only if explicitly in input (OP-42)

Decision rules

Input mentions	Set
Deploy prod / delete / external email	risk_level ≥ 3
Ticket key in text	ticket_key field
Multi-step work	Stop here — planner decomposes (open-orchestrator-plan)

Pitfalls

• Inventing ticket IDs not in the sentence
• Decomposing steps (planner's job)
• Markdown-wrapped JSON

Verification

• Output parses as openorchestrator.normalized-goal.v1
• approval_required true when risk ≥ 3
• correlation_id propagated on next hop