@dexto/agent-config

Schema + resolver utilities for turning an agent YAML config into concrete DI surfaces using images.

This package is product-layer glue: core (@dexto/core) stays DI-friendly, and hosts (CLI/server/apps) use @dexto/agent-config to:

• validate agent config (AgentConfigSchema)
• load an image module (loadImage())
• merge image defaults into raw config (applyImageDefaults())
• resolve factories from the image into concrete instances (resolveServicesFromConfig())
• convert the result into DextoAgentOptions (toDextoAgentOptions())

Quick example (apps)

import { DextoAgent } from '@dexto/core';
import { loadAgentConfig } from '@dexto/agent-management';
import {
  AgentConfigSchema,
  applyImageDefaults,
  loadImage,
  resolveServicesFromConfig,
  setImageImporter,
  toDextoAgentOptions,
} from '@dexto/agent-config';

// Under pnpm (strict dependency boundaries), configure image importing at the host entrypoint.
setImageImporter((specifier) => import(specifier));

const raw = await loadAgentConfig('./agents/my-agent.yml');
const image = await loadImage(raw.image ?? '@dexto/image-local');
const merged = applyImageDefaults(raw, image.defaults);

const config = AgentConfigSchema.parse(merged);
const services = await resolveServicesFromConfig(config, image);

const agent = new DextoAgent(toDextoAgentOptions({ config, services }));
await agent.start();

Images (no registries)

Images are typed modules (DextoImage) that export plain Record<string, Factory> maps. Resolver logic does property access by config type (e.g., image.tools[entry.type]) — there are no global registries or side effects.