I build open-source tools for production-grade AI agent workflows:
generate trusted context → expose it as tools → verify agent behavior → package reliable skills.

The focus is not prompt-and-pray demos.
The focus is source-grounded context, MCP tool integration, evals, replayable failures, and verification-gated agent skills.

Source Code Repository
        │
        ▼
   Open Mind ───────────────────── verification-first codebase analysis
        │                           Python · standalone
        ▼
 .openmind artifacts ───────────── versioned schema contract
        │                           manifest.json · file:line evidence
        ▼
 open-mind-mcp-server ──────────── MCP tool integration layer
        │                           TypeScript · standalone
        ▼
 Claude / Cursor / AI agents ──── structured tools, cited answers, honest refusals
        │
        ▼
 agent-skill-verification-template
        │                           evals · validators · replay artifacts · quality gates
        ▼
 agent-skill-forge ─────────────── spec-driven skill generation and packaging pipeline

Each project runs standalone. They integrate through narrow, versioned contracts — no monorepo, no hidden coupling.

Verification-first codebase context for AI agents.

Open Mind turns local repositories into deterministic, source-traceable knowledge artifacts with file:line evidence.
It is designed for agents that need to understand codebases without relying on model memory or unsupported summaries.

Core ideas: source grounding, deterministic extraction, codebase context engineering, honest refusal when evidence is missing.

MCP tools for source-grounded codebase understanding.

This server loads .openmind artifacts and exposes them to MCP-compatible clients such as Claude Code, Claude Desktop, and Cursor.

It provides structured tools for:

• searching codebase context
• retrieving symbol evidence
• explaining architecture components
• validating claims against source references
• listing available artifacts

Core ideas: MCP integration, tool contracts, structured JSON outputs, agent-callable evidence.

A quality gate for AI agent skills.

This project treats agent skills as production components.
It provides an offline eval harness, validators, replay artifacts, metrics, static reports, and release gates.

Core ideas: evals, golden cases, negative cases, citation validation, replayable failures, model-independent skill contracts.

A spec-driven pipeline for generating, testing, repairing, and packaging AI agent skills.

agent-skill-forge turns a structured skill requirement into a verified skill package:

skill requirement
  → skill spec
  → generated skill files
  → eval cases
  → verification
  → repair loop
  → quality gate
  → installable package

Core ideas: agent skill generation, skill SDLC, verification-gated packaging, model-agnostic skill development.

• Evidence over fluency
  Important answers should point back to real sources or clearly say that evidence is missing.

• Specs before generation
  AI is an execution accelerator, not the source of truth. Requirements, contracts, tests, and quality gates come first.

• Contracts over coupling
  Tools and agents should integrate through explicit schemas, not hidden assumptions.

• Measured, not asserted
  Reliability should be demonstrated through evals, reports, replay artifacts, and failure analysis.

• Production-minded AI
  Agent workflows need observability, permission boundaries, partial failure handling, and release gates.

AI agents · MCP · Claude skills · agent skills · LLM evals · tool calling · context engineering · verification-first systems · source-grounded AI · production AI workflows

MIT-licensed. Issues and PRs welcome.