Grillo — History Evaluator (helper)

Part of the G.R.I.L.L.O. background subsystem.

Purpose

A non-LLM helper — not a beat. It formats the last N chat messages into a concise reflection prompt that several beats reuse, so each beat doesn’t have to reimplement history formatting.

What it is

• A PluginBase (no LLM) with its own get_metadata() for the WebUI.

• Has no BEAT_TYPE: it is never weighted-selected by the scheduler; the Grillo core (grillo_impl.py) fetches it explicitly from the plugin registry and calls it while building beat prompts.

How it works

Given an interface_path and a message count, it reads recent chat history and returns a compact, prompt-ready summary of the conversation. Purely deterministic formatting — no model call.

Configuration

No dedicated config keys. Behaviour is driven entirely by the caller (which beat requested the reflection and for how many messages). See the G.R.I.L.L.O. guide.