The workbench
Platform work and weekend experiments.
Some of this ships for thousands of developers. Some of it was a Saturday on a Mac. Both belong here.
In progress & shipping
Everything on the bench
Ordered by how actively I'm touching it. Open a card for the honest write-up, including the caveats.
- CLI · Agent readiness
Agentlint
Lighthouse for agents. A deterministic website scanner that tells you whether agents can find, understand, and use a site, and proves every conclusion with evidence.
Active · @timbenniks/agentlint
- SDK · TypeScript
Contentstack Platform SDK
One TypeScript surface for Contentstack (CMA, OAuth, webhooks, image transforms, and generative AI) so you stop stitching four libraries together before you write the actual product.
Active · @timbenniks/contentstack-platform-sdk
- OSS · Nuxt module
Nuxt Contentstack
The Nuxt module I wish Contentstack shipped: typed composables, live preview, visual builder, personalization, and image optimization in one install.
OSS · nuxt-contentstack on npm
- OSS · Models as code
Contentstack Stacksmith
Content types as TypeScript you can import, compile, diff, document, type-generate, and safely promote, so the UI is no longer the source of truth.
OSS · @timbenniks/contentstack-stacksmith
- OSS · WebMCP
Contentstack WebMCP
Framework-agnostic tools that expose published Contentstack content to in-tab agents via document.modelContext. Vanilla TypeScript first, React and Next adapters if you want them.
Experimental · @timbenniks/contentstack-webmcp
- MCP · Agent skills
Contentstack MCP Hub
A hub of agent skills with RBAC, so an agent acting for a user can only reach what that user is allowed to touch, and product managers can propose the next skill without writing it.
Active · Internal + open contributions
- macOS app · MCP
Open Loops
A local-first macOS tracker for work you hand to coding agents. SQLite on your machine, git-aware tasks, and a full backlog API over MCP.
Open source · macOS · MIT
- MCP · Local-first
Turbo Relay
A local-first intelligence layer for sanctioned AI agents. Agents write evidence-backed knowledge through MCP; humans review, approve, and promote it.
Open source · macOS · MCP
- TUI · Agent orchestration
Dex
Local-first AI engineering orchestration for the terminal. Agents write the code. Dex runs the rest of the loop: ticket, worktree, tests, PR, review, and the human gate before merge.
Active · local-first · Bun
- macOS app · Skills
Loadout
A native macOS control plane for Skills and MCP servers across Claude Code, Codex, and Cursor. It does not own your loadout. It shows you what is actually installed.
Early · macOS 15 · Swift
- macOS app · Local-first
ButterKeys
A menu bar app that fixes the typos your hands already know how to make (teh, writign, int he) on the machine, with no account and no network.
macOS 14 · local-first
- OSS · Agent workflow
Cursor Rules
A reusable Cursor ruleset and command kit so the agent behaves like a senior engineer on a SaaS project: constrained, repeatable, and boring in the ways that matter.
Open source · MIT
- OSS · CLI
Contentstack OpenAPI Plugin
A csdx plugin that introspects a Contentstack stack and writes a valid OpenAPI 3.1 spec for the Delivery or Management API, including schemas, regions, modular blocks, and security schemes.
OSS · @contentstack/cli-plugin-openapi
- Experiment
DoorBell DevRel
AI-generated developer videos at scale, with almost no production code. A live test of what adaptive content looks like when making one more video costs almost nothing.
Live · doorbell-devrel.timbenniks.dev
- Workgroup
MACH Alliance Enterprise AI Agents
Co-chairing the workgroup that is trying to define how composable architecture and AI agents get adopted when the buyer has real governance, not a weekend demo.
Ongoing · MACH Tech Council
Want to collaborate, contribute, or drag a weekend experiment into production? Write me.