local-runtimes
Use when choosing an optional local inference runtime for a narrow mechanical or offline job. One table routes to OmniRoute, FreeToken, Colibri, a loopback server, or the hardware probe.
- License 许可
- MIT
- 最近更新
- 2026-10-20
- Stars
- ★ 31
- Forks
- 4
01 前言 Front Matter
- name
- local-runtimes
- license
- MIT
- description
- Use when choosing an optional local inference runtime for a narrow mechanical or offline job. One table routes to OmniRoute, FreeToken, Colibri, a loopback server, or the hardware probe.
02 资源文件 6
- 参考文档references/colibri.md1.3 KB
- 参考文档references/freetoken.md1.3 KB
- 参考文档references/local-coding-delegate.md1.5 KB
- 参考文档references/local-openai-compat.md4.7 KB
- 参考文档references/loopback-runtime-contract.md2.9 KB
- 参考文档references/omniroute.md13.6 KB
03 SKILL.md 正文
Local runtimes
Cloud implementers write code directly. A local runtime is optional and only pays for narrow mechanical or offline jobs; never let it review its own diff. Pick one row, then open only that route.
| Need | Route | Open |
|---|---|---|
| Which model size fits this machine | hardware probe | local-coding-delegate |
| Already READY loopback server (Ollama, LM Studio, llamacpp) | loopback | local-openai-compat |
| Standalone FreeToken process | FreeToken | freetoken |
| Frontier MoE through disk/RAM/VRAM tiers | Colibri | colibri |
| OmniRoute runner with delegate continuity | OmniRoute | omniroute |
| Local agents (OpenCode / OSS agency) | agency | local-agency |
FreeToken and Colibri share the loopback runtime contract.
Rules:
- A runtime is used only when its READY probe passes; otherwise stay on the
cloud implementer and record
local: skipped(<reason>)once. - One-shot jobs only: one short spec in, one digest out. No parent relay.
- Avoided-spend metering and USD/EGP lines are required only when a local inference channel was actually used (process owner).
- Descriptions, paths and host exposure live in the catalog.
04 同仓库更多 Skills
ponytail
Lazy senior dev mode: the smallest change that fully solves the task, and a reply a busy human understands in one read. Use on any coding task (writing, fixing, refactoring, reviewing, choosing dependencies) and when the user says "ponytail", "be lazy", "simplest solution", "yagni", or complains about over-engineering or bloat. Levels: lite, full (default), ultra.
icm-architect
Advisory ICM workspace architect (folder structure as agent architecture). Use only when the task is ICM, workspace structure, or 'ICM this' design work.
chaos-engine
Load the canonical installed ChaosEngine before every task.
ultracave
Caveman at maximum compression: fragments, one word when one word is enough, each fact once. Invoke only with /ultracave or /caveman ultra. Stays on until "stop caveman" or "normal mode".
caveman
Ultra-compressed communication companion. Use on every reply at ultra intensity; mandatory on the implementation path. Stop only with 'stop caveman' or 'normal mode'.
chaos-engine
Load the canonical installed ChaosEngine before every task.