@arnilo/prism-caveman

Optional Caveman behavior integration for Prism (0.0.23).

Importing is inert: no skills, commands, timers, or network until the host loads the extension via kernel.load([createCavemanExtension(...)]).

Requirements

• A local checkout of juliusbrussee/caveman (not published on npm).
• Host-supplied upstreamPath pointing at that checkout (must contain a skills/ directory).
• Session callbacks (appendEntry, getEntries) for level persistence — same pattern as observational memory attach.
• Peer @arnilo/[email protected].

Quick start

import { createCavemanExtension } from "@arnilo/prism-caveman";
import { createExtensionKernel, createLoadSkillTool, createLoadedSkillSet, createSkillRegistry } from "@arnilo/prism";

const caveman = createCavemanExtension({
  upstreamPath: "/path/to/caveman",
  defaultLevel: "full",
  appendEntry: (entry, options) => store.append(entry, options),
  getEntries: () => session.entries(),
});

await kernel.load([caveman]);
// Select instruction injector "caveman-mode" on runs that should receive level slices.
// Register createLoadSkillTool + skillsDisclosure: "progressive" for catalog-only skill bodies.

Registers the required seven upstream skills plus any extra skills/*/SKILL.md. Level command + caveman-mode injector (filtered skills/caveman/SKILL.md) + session caveman-level. Caveman 2 engine files are ignored.

Full API docs: docs/caveman.md. Demo: node examples/caveman-ponytail.ts.