Agents

A subagent is a one-shot, role-prompted invocation of the host runtime's "spawn a subordinate agent" capability. Agents do not run in a daemon; they spawn, execute, and exit. Workflows orchestrate, agents read, write, and reason.

agents/ ships 17 spawnable agents and 4 audit-surface modules, 21 files total. Modules carry module: true in their frontmatter and are loaded via Read by their parent spawnable agent; lib/agents.cjs::loadAgent rejects them for direct spawning. The modules are np-critic-style, np-critic-tests, np-critic-acceptance, and np-critic-economy, all loaded by np-critic per the Single-Critic Revision (ADR-0010 §2026-05-05). The first three load on every spawn; np-critic-economy loads only when agents.economy resolves to full or ultra (see the Economy axis). np-researcher-reconciler (Stage 2 of the researcher swarm, ADR-0018) joined the spawnable set on 2026-05-12. np-learnings-extractor (the background continuous-learning observer, ADR-0010) joined on 2026-06-14. np-simplifier, the read-only economy reviewer behind /np:simplify-review, joined on 2026-06-25. np-task-architect (the per-task architecture step) and np-test-writer (the per-task TDD step) joined the round-1 Nubosloop on 2026-06-30 — both config-gated (agents.architect / agents.test_writer, default on) per ADR-0023.

The orchestration graph is exactly one level deep: agents may not spawn other agents.

Where to look

Topic	Page
Full per-agent details (tier, tools, spawn-source, handoff protocol)	Catalog
The frontmatter contract (required fields, forbidden fields, tier enum)	Frontmatter Schema
Why no model / hooks field, what tier means, the mandatory-initial-read invariant	Concepts → Agents
Tier × Profile model resolution	Agent Frontmatter Schema § Model resolution