agent-harness-kit

A provider-agnostic scaffolding kit for running structured multi-agent workflows in your codebase.

★ 1829 forksTypeScriptApache-2.0enmanuelmag

Provider-agnostic CLI that scaffolds a structured multi-agent coding harness into your repo: shared task backlog, role-based agents, SQLite audit trail, health gate, and a local MCP server that Claude Code, OpenCode, Codex, or Grok can connect to. Stored locally, no cloud or API keys beyond your existing AI tool.

Use Cases

Scaffold a structured 5-role agent workflow (Lead/Explorer/Consultant/Builder/Reviewer) into a repoCoordinate multiple coding agents via an atomic SQLite-backed task backlogLog every agent action, file touched and blocker for audit and handoffExpose harness tools to Claude Code, OpenCode, Codex CLI or Grok via a local stdio MCP serverEnforce a health gate before agents start or close a taskSwitch AI providers without losing task history or workflow configSearch project docs from within an agent before writing codeTurn feature/fix requests into reviewable spec drafts in docs/specsRun a local dashboard to inspect harness status and activityShare one global harness across all projects via ~/.claude or ~/.config/opencodeSwap SQLite for PostgreSQL or MySQL with a single config linePin reproducible agent workflow behavior across team and CI

Built With

Language: TypeScript
Frameworks: Model Context Protocol SDK · Hono · Drizzle ORM · Commander · Valibot · Vite · tsup · better-sqlite3 · OpenTelemetry · pnpm

agent-harness-kit

Use Cases

Built With

Tags