Local-first
Orchestration you can inspect
Routing, fallback logic, caching, and safety checks run locally. You see exactly what happens and why.
SkillForge exists because real developer workflows already span multiple models and clients — and the skills you write for one should not be trapped inside it.
Most developers now use more than one coding agent:
The problem is not access. It is coordination. Without a shared layer, teams end up with duplicated prompts, inconsistent routing, no cost controls, and skills that get trapped inside whichever client happened to be used first.
SkillForge fixes that.
Local-first
Routing, fallback logic, caching, and safety checks run locally. You see exactly what happens and why.
Portable
The pack format is designed to travel. Write once, use across Codex, Claude Code, Kimi Code, and any MCP-capable client.
Transparent
Budget pressure, fallback behavior, safety constraints, and routing decisions are explicit operating choices — not hidden side effects.
Practical
Standard stdio installs. Straightforward config. Exportable packs. Working software matters more than demo theatrics.
Run one shared orchestration layer across multiple coding clients with cost-aware routing, semantic caching, safety guardrails, budget controls, and a Streamlit dashboard.
Browse, inspect, and export 302 standalone skill packs — even if you never run the runtime. Every pack is a real folder with SKILL.md, skill.yaml, marketplace.yaml, and README.md.
SkillForge is a strong fit if you: