Agents

Agents are the persistent, autonomous actors that perform work in matic. Unlike stateless tools invoked per-prompt, agents have durable identities, accumulate experience across engagements, own capability profiles that sharpen over time, and hold positions within teams. This section covers everything needed to understand what agents are, how they are composed, how they develop, and how they coordinate.

Archetypes

Reusable specification templates — blueprints from which agents are spawned. Covers authoring, the capture workflow that promotes an experienced agent's refined profile into a new Archetype version, and deprecation.

Collaboration

The seven named Collab Modes (handoff, delegate, consult, review, pair, shadow, escalate) and the agent.message Signal type that carries work between actors. Covers delivery semantics, mid-execution routing, and how Helpers participate through their parent agent's communication boundary.

Identity

How agents are addressed and distinguished: handle as the sole identifier, the singleton invariant (at most one live runtime instance per handle), @mention routing, and how identity remains stable across Agent Runtime swaps.

Lifecycle

The full agent state machine from Spawned through Onboarding, Available, Engaged, Learning, Assessment, and Idle, plus the Suspended and Decommissioned paths. Covers the mandatory five-step onboarding sequence, the Learning state that closes every engagement, and the decommission flow that routes experience back into the org.

Memory & Experience

Three distinct persistence layers — Experience (append-only engagement log), Memory (curated mutable context), and Library (immutable reference store) — and how agents read and write each during the work lifecycle. Also covers Skills as structured capability specifications that give Tools operational context.

Probes

Lightweight, side-effect-free heuristics that describe an agent's capability, confidence, bandwidth, availability, and exclusions. Covers the evaluation DSL, the grounding basis requirement in confidence Probes, and how agents inherit Probes from their Archetype and refine them through Experience.

Staffing

The process by which matic composes a Team for a Project: competence-based candidate matching via Probes, Squad Templates for repeatable compositions, Auto Matic's spawning authority and criteria, and the human approval gate before engagement begins.