Skip to content

wangmiaozero/laya-router-skill

laya-router

Verified

Cross-platform local decision routing for AI coding agents using Laya and Laya-MLX. Use for fast task classification, complexity estimation, model escalation hints, tool-use routing, and lightweight risk assessment. Advisory only.

FileKindSize
SKILL.mdskill md1.6 KB
agents/openai.yamlconfig219 B
assets/icon.svgother2.9 KB
references/AGENTS.mdmarkdown1.1 KB
references/AGENT_E2E.mdmarkdown8.2 KB
references/ARCHITECTURE.mdmarkdown1.2 KB
references/BACKENDS.mdmarkdown967 B
references/RC_VALIDATION.mdmarkdown5.0 KB
scripts/decide.pyscript561 B
scripts/decide.shexecutablescript64 B
scripts/healthcheck.pyscript3.3 KB
scripts/healthcheck.shexecutablescript69 B
scripts/install.ps1script226 B
scripts/install.pyscript12.3 KB
scripts/install.shexecutablescript65 B
scripts/uninstall.pyscript6.5 KB
scripts/uninstall.shexecutablescript73 B

references/ARCHITECTURE.md

View on GitHub
# Architecture

```text
AI agents (ChatGPT Desktop (Codex), Codex CLI, Claude Code, OpenCode, Pi)
  -> Agent Skill -> laya-router CLI
                  -> optional MCP adapter
                  -> RouterCore -> LayaBackend
                                 -> laya-mlx / MLX (Apple Silicon)
                                 -> upstream laya / PyTorch (other platforms)
```

The CLI is the stable entry point. MCP exposes `laya_decide`, `laya_health`, and `laya_info` over stdio and shares the core. The runtime never executes model output. All errors yield `status=unavailable`, `advisory=true`, and `fail_open=true`. In `backend=auto`, Apple Silicon tries MLX then PyTorch. An explicit backend never silently changes.

The upstream Router handles language and checkpoint selection. Its model instances are lazy and retained in the long-running process, with at most two loaded at once. A separate CLI process has a separate lifetime; use MCP for hot repeated requests.

The Python environment and config live in a platformdirs user data directory. Agent skill links point to a separate copied skill tree. The installer records ownership in `install-manifest.json`; uninstall consults that file before deletion.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall