SkillForge: open-source tool that proves Claude Code skills work via Red-Green testing
tom_doerr · x · 2026-08-19
SkillForge v6 (854 stars on GitHub) is an open-source, evidence-driven skill creator for Claude Code and Codex. Its core principle: skill quality is a property of behavior, not documents — a skill is done only when a fresh agent demonstrably performs better with it than without.
Key features:
- Red-Green execution pipeline: runs a baseline without the skill (RED) before writing, then with-skill runs (GREEN) to verify real impact
- Per-skill regression evals and ecosystem health maintenance
- Cross-runtime compilation between Claude Code and Codex, an ecosystem doctor, and an opt-in proactive advisor
v6 is a ground-up rework after a deep external audit, dropping v5's "4-agent unanimous synthesis panel" in favor of actually executing skills for verification.
More from coding & agent
- Automation platform Relay.app shuts down; free users' data deleted Aug 15, 2026 — HaktanSuren · 2026-08-19
- RepoRelay: Secure MCP bridge to connect ChatGPT to local repos — Lucaslogged · 2026-08-19
- Agent-Driven Three.js Visualizations with Real Physics and Good UI/UX — doodlestein · 2026-08-19
- Volkern MCP Server Integrates CRM with AI Agents — modelcontextprotocol · 2026-08-19
- Gate Info MCP for Crypto Discovery and On-chain Data — modelcontextprotocol · 2026-08-19
- Hosted OAuth MCP Server for Task Management and RFC 9728 Gotchas — Mean-Papaya3532 · 2026-08-19