@adhd/agent-engine-compiler

Compiles agent configurations from the registry into runtime-ready manifests. Resolves component compositions, tool bindings, policy constraints, and model specifications. Part of the agent registry family.

Status: Shipped v2.1.2
Platform: platform:node (runs in Node/CLI, not browser)
Consumers: @adhd/agent-mcp, @adhd/dispatch-cli

What it does

• Body Resolution — Assembles agent components from the registry composition store into a complete prompt body
• Tool Resolution — Resolves tool aliases and platform-specific bindings (HTTP, MCP, CLI, etc.)
• Model Resolution — Looks up model capabilities and configurations
• Policy Constraint Resolution — Evaluates effective authorization policies
• Output Emission — Compiles agent definitions into YAML frontmatter, JSON objects, or Markdown documents
• Database Migrations — Runs schema setup via runMigrationsOn(db)

Usage

First, resolve the shared registry database path:

import { resolveRegistryDbPath } from '@adhd/agent-core-env';
import Database from 'better-sqlite3';
import { runMigrationsOn, compileAgent } from '@adhd/agent-engine-compiler';

const dbPath = resolveRegistryDbPath(); // e.g., ~/.adhd/agent-registry/production/data/registry.db
const db = new Database(dbPath);

// Run migrations
await runMigrationsOn(db);

// Compile an agent from the registry
const compiled = await compileAgent(db, {
  agentId: 'my-agent',
  platform: 'mcp', // or 'http', 'cli', 'json'
  models: modelStore,
  tools: toolStore,
  policies: policyStore,
  composition: compositionStore,
});

// Emit as YAML or JSON
const yaml = emitYamlFrontmatter({
  agentId: compiled.agentId,
  tools: compiled.tools,
  modelAlias: compiled.modelAlias,
  policyConstraints: compiled.policyConstraints,
});

Key exports

| Export | Purpose | |--------|---------| | resolveBody(db, ...) | Assemble component composition into prompt body | | resolveTools(db, ...) | Resolve tools and platform bindings | | resolveModel(db, ...) | Look up model and capabilities | | resolvePolicyConstraints(db, ...) | Resolve effective policies | | compileAgent(db, input) | Complete compilation orchestrator | | emitYamlFrontmatter(input) | Emit as YAML frontmatter + body | | emitJsonObject(input) | Emit as JSON (e.g., for REST API) | | runMigrationsOn(db) | Initialize database schema |

Architecture

• Tier: engine (depends on base + core + stores)
• Consumption: Reads from all registry family stores (prompts, tools, policy, provider)
• No import-time side effects — DB path and connection are external (via agent-core-env)
• Part of agent registry family: agent-store-prompts + agent-store-tools + agent-core-policy + agent-core-provider + agent-engine-compiler

See /entrypoint/agent-mcp/ for the agent MCP server (primary consumer).