- Initial release — live demo and downloadable, runnable foundation.
Agent Workbench
A workspace for prompts, tasks, artifacts, and coding-agent runs.
Agent requests, decisions, artifacts, and human review in one stream.
Reusable recipes and context bundles for repeatable agent work.
Generated outputs with approval states and follow-up tasks.
Overview
Agent Workbench is built for the AI coding era: a practical foundation for teams that want to capture prompts, run context, generated files, review notes, and follow-up tasks. It gives agents a predictable domain model while giving humans a clean audit trail.
Features
- Agent run timelines
- Prompt and context libraries
- Generated artifact review
- Human approval states
- Provider-agnostic model boundaries
Make it yours in three steps
Download it
One click, one zip, the whole project. Free, MIT licensed, yours to keep.
Download Agent WorkbenchRun it
Unzip it, then run this in the folder. It works immediately — no database, no accounts, no configuration.
$ npm install
$ npm run devHand it to your coding agent
The download ships with CLAUDE.md, AGENTS.md, and Cursor rules, so your agent understands the project the moment it opens the folder.
$ cd agent-workbench
$ claudeClaude Code reads the included CLAUDE.md automatically — it already knows the project.
A good first prompt:
“Give me a quick tour of this project, then help me make it mine.”Agent-ready information
The repository is organized around explicit agent concepts and ships with recipes for adding providers, artifacts, and review workflows.
Changelog
Related foundations
Explore nearby starting points before you commit to a fork.
/education / updated Jul 1, 2026
Knowledge Base
Docs, collections, search, and publishing workflows.
/productivity / updated Jun 29, 2026
Personal Operating System
Tasks, notes, routines, and planning in one owned workspace.
/developer-tools / updated Jul 1, 2026
Marketplace Foundation
Listings, discovery, detail pages, and trust signals.