Probes

Probes are lightweight, side-effect-free heuristics that agents author and maintain to describe their own capability, confidence, availability, and limits. Rather than invoking an agent's full reasoning to determine suitability, matic reads Probes as plain data at routing and staffing time — making selection decisions in milliseconds with zero GenAI calls. Agents inherit an initial Probe set from their spawned Archetype and refine it through the Learning phase as Experience sharpens their self-model.

Capability

What the agent can do and under what constraints, expressed as matchable task-class patterns inherited from the Archetype and extended through Experience.

Bandwidth

The agent's current and projected capacity given its active Work Pile and budget envelope, surfaced to the routing layer to prevent over-assignment.

Availability

The agent's current scheduling state — whether it can accept new work immediately — distinct from bandwidth and updated as the agent transitions between lifecycle states.

Confidence and Grounding

The agent's self-assessed confidence for a domain or task class, including the grounding basis it treats as authoritative; low confidence with weak grounding triggers automatic HITL escalation before execution.

Exclusion

Hard limits on what the agent will not do, routing work away from the agent at the dispatch layer before any engagement begins.

Evaluation Contract

The DSL and runtime rules governing how Probe expressions are parsed and evaluated — strictly side-effect-free and authored in plain markdown so agents can maintain them without specialized tooling.

Authoring and Refinement

How agents write and evolve their own Probes: the Learning phase workflow, proposing provisional updates, Archetype inheritance constraints, and when Experience evidence is required to extend or narrow a base Probe.