Enchiridion
Labs

Fluent infrastructure for AI agents that actually work.

Ergonomic and composable foundational tools that let AI agents access data, follow procedures, and deliver results through generated interfaces.

July 2026 SkillOps — the missing operating layer for governing, distributing, measuring, and improving agent skills → July 2026 HarnessC is open source — specialized AI harnesses for lower cost and higher task performance → May 2026 Proceda Voice — SOP-driven voice agents with bounded procedural attention → May 2026 Specialized Harness Engineering — principles for turning LLMs into reliable narrow-task workers → March 2026 Proceda achieves SOTA on SOP-Bench — beating Claude Opus baselines with lightweight models →
01 — The Thesis

Three Abstractions That Make
AI Agents Effective

Three standardized capabilities have recently emerged that, when combined, enable AI agents to execute complete workflows from start to finish.

Abstraction 01

MCP

Model Context Protocol

Standardized access to data and APIs. A universal adapter between agents and the world's services — no bespoke integrations required.

Abstraction 02

Skills

Procedural Knowledge

Encoded instructions, SOPs, and domain knowledge. Agents follow known procedures reliably instead of improvising from scratch.

Abstraction 03

Generative UI

Dynamic Interfaces

AI-generated interfaces for human consumption. Every agent result is unique, so the UI must be generated to fit the specific output.

Skills encode the what. MCP handles the how. Generative UI delivers the output.

Read the full thesis →

02 — Products

Products

Three products spanning governed agent skills, procedural execution, and composable MCP infrastructure.

skillops.app

SkillOps

Govern the skills your agents depend on.

The operating layer for authoring, approving, distributing, observing, and improving agent skills—with analytics that distinguish usage from efficacy.

  • Governed skill registry and version history
  • Review and approval workflows
  • Distribution through a managed MCP gateway
  • Usage and outcome analytics
  • Vendor-neutral skill lifecycle
Explore SkillOps →
# Governed lifecycle author ↓ approve ↓ distribute ↓ observe ↓ measure ↓ improve usage: who runs it? efficacy: does it work?
sop.run

Proceda

You documented it. Now let agents automate it.

Transform standard operating procedures written in natural language into autonomous AI workflows with human-in-the-loop approval gates. Your SOPs become executable agents.

  • Natural language workflow execution
  • MCP tool integration
  • Human approval checkpoints
  • Multi-LLM support with BYOK
  • Enterprise SSO
Visit sop.run →
# skill.md name: Invoice Processing tools: [email, database, slack] Steps: 1. Read incoming invoices from email 2. Extract vendor, amount, due date 3. Cross-reference with PO database [APPROVAL REQUIRED] 4. Post to accounting system 5. Notify team on Slack
npm package

mcpknife

A Swiss Army knife for MCP servers.

Generate, transform, and add UIs to MCP servers with Unix pipe composition. Go from “I have an API” to “I have an MCP app with interactive UI” in one command.

  • boot — Generate servers from natural language
  • mod — Reshape tools via natural language
  • ui — Auto-generate interactive UIs
  • Unix pipe composition
  • Zero-code, natural language only
View on GitHub →
# Generate + transform + add UI in one pipe mcpknife boot --prompt "GitHub API tools" \ | mcpknife mod --prompt "combine issues and pulls into get_activity" \ | mcpknife ui # Result: MCP server with interactive UI # ready for ChatGPT or Claude
03 — The Stack

How It Fits Together

The infrastructure stack connects agent interfaces, procedural execution, and MCP access to underlying systems.

User / ChatGPT / Claude
Generative UI
Dynamic interfaces for humans
Skills Engine
Procedural knowledge execution
Proceda
MCP Infrastructure
Server generation & transformation
mcpknife
Raw APIs / Databases / Services
04 — Ideas

Research & Writing

The intellectual foundation behind the tools.

Open Source

HarnessC Is Open Source

A compiler-shaped answer to AI token costs: turn task specifications into narrow harnesses that can run cheaper models and perform better on the target workflow.

Read →
Thesis

The Three Abstractions That Make AI Agents Real

Three standardized capabilities — MCP, Skills, and Generative UI — click together to enable agents that execute complete, end-to-end business processes.

Read →
Platform

Birth of a New Platform

The ChatGPT App Store as the next major platform opportunity, with 800 million weekly active users and a nascent developer ecosystem.

Read →
Framework

Read-Cognify-Write

Knowledge work decomposed into atomic loops. AI as a compiler that optimizes, parallelizes, and eliminates unnecessary steps in these loops.

Read →
Voice Agents

Proceda Voice: SOP-Driven Voice Agents

A technical article on natural-language SOPs, windowed slot frontiers, and deterministic state for reliable low-latency voice agents.

Read →
Principles

Specialized Harness Engineering

A field guide for turning open-ended LLM capability into constrained, measurable, task-specific execution environments.

Read →
05 — About

About the Lab

Enchiridion — from the Greek ἐγχειρίδιον, meaning “handbook” — was Epictetus’s compact guide to what matters. The name reflects the same principle applied to AI agents: reliable execution comes from well-structured procedures, not improvisation.

Enchiridion Labs builds the infrastructure layer of the AI agent ecosystem. The thesis is that agents need three standardized capabilities to work in practice — data access (MCP), procedural knowledge (Skills), and dynamic interfaces (Generative UI). Each of these has recently become a real, shipping standard. The tools sit at this intersection, creating the infrastructure that makes agent-driven automation real.