Toolshed
An agent capability and operations layer — a library of reusable skills, plus executable services, MCP integrations, scheduled workflows, approvals, monitoring, and recoverable deployment.
Overview
Toolshed is the capability and operating layer I built around the agent runtimes I use. It packages a library of reusable skills in a portable format that works across agent runtimes, together with executable services, MCP integrations, scheduled workflows, shared state, approval boundaries, monitoring, and multi-host deployment. None of it was planned. I built each skill for work I was already doing, kept the ones that stayed useful, and they formed these buckets naturally as the pile grew: prediction, commerce, infrastructure, content, voice.
Constraint
An agent demo can call a tool once. An operating capability platform has to make hundreds of heterogeneous workflows discoverable, repeatable, permission-aware, observable, and recoverable across machines without letting automation outrun ownership or safety boundaries.
Contribution
Built a portable skills and services layer spanning research, commerce, prediction, browser automation, voice, infrastructure, and content. MCP tools, scheduled execution, shared state, approval gates, alert routing, and deployment workflows run across Mac and NVIDIA Jetson hosts. Health checks, circuit breakers, canaries, drift tripwires, reconciliation, audit trails, and rollback controls keep operations bounded.
Architecture
Input
System
Outcome
Capabilities
Stack
Next
Want something like it built for your team?
Start a conversation→