using-nous

Using NOUS

Core Principle

Route first, then load exactly one planning personality.

When To Use

• The user invokes nous:*.
• The user asks for a plan from a domain specialist.
• The user asks for UI, ultra UI, database, backend, QA, CEO/product, security, performance, release/deploy/infra planning, an ultraplan, plan execution, quick execution, TDD, debugging, code review, branch finishing, parallel agents, verification, or workflow resource ingestion.
• The user asks for a natural-language plan that maps cleanly to one NOUS planner.

When Not To Use

• The user asks a simple factual question.
• The user asks you to directly implement a small change and no specialist plan is needed.
• The request is pure conversation.

Router Order

1. Check explicit commands first in core/router/triggers.json, including planners, execution commands, verification, TDD, debugging, code review, branch finishing, parallel agents, nous:write-skill, and nous:ingest-workflow-resource.
2. If no command is present, match natural-language triggers from core/router/triggers.json.
3. If exactly one planner matches, load only that planner.
4. If multiple planners match, ask one short clarification question.
5. If no planner matches, say NOUS is not needed and continue normally.

HARD-GATE: Single Active Personality

Do not blend planner personalities unless the user explicitly asks for a multi-role review.

write-a-skill, ingest-workflow-resource, and nous-execute-plan are meta-skills, not planner personalities. They must still avoid loading unrelated planner personalities unless the task explicitly needs them.

Execution Routing Rule

If routing to nous-execute-plan, the executor must check for an existing plan before doing any implementation. If no plan exists, it must stop and propose the correct planning command.

Forbidden rationalizations:

• "A complete plan should include everything."
• "The adjacent concern is probably useful."
• "The implementation needs all domains anyway."

Planning Contract

Every planner must use this output shape:

## Objective
## Domain Analysis
## Specialized Tasks
## Dependencies
## Risks
## Validation Requirements
## Success Metrics
## Not In Scope

Inline Self-Review

Before returning a plan, check:

• Did I load only one planner?
• Did I avoid adjacent-domain tasks?
• Did I include Not In Scope?
• Did I mark unknowns instead of inventing facts?
• Did I keep the plan domain-specific rather than generic?

Output Contract

If routing succeeds, respond with the selected planner's plan. If ambiguous, ask one clarification question with the likely planners.