智图云仓 Token 路由
v1.0.1智能模型路由 - 根据任务是否需要工具调用自动选择模型。需要工具(搜索/浏览器/文件/天气/股票等)使用百炼付费模型，纯聊天使用OpenRouter免费模型。V3.1.0新增route_by_tool_required()函数。
⭐ 0· 59·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
Capability signals
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (token optimizer, lazy loading, model routing, heartbeat tuning, budget tracking) align with the included scripts and docs. The code reads/writes only OpenClaw workspace files and performs text classification and local state tracking — appropriate for the stated goals.
Instruction Scope
SKILL.md runtime instructions (run the bundled Python scripts, copy heartbeat template, generate AGENTS.md) map to the bundled scripts. The instructions reference only local workspace paths (~/.openclaw/workspace) and do not instruct automatic network calls or harvesting unrelated system data.
Install Mechanism
No remote install/downloads are declared; the package is instruction-with-local-scripts only. The provided shell wrapper (optimize.sh) is a convenience that invokes bundled python scripts. There are no URLs used to fetch or extract code at install time in this bundle.
Credentials
The skill declares no required environment variables (none listed). Some scripts (model_router.py) detect optional provider API keys from the environment (ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_API_KEY, OPENROUTER_API_KEY) to choose a provider if present — this is reasonable and optional. Be aware: providing those keys to your agent environment would enable the skill to choose models that rely on them; the skill itself does not appear to transmit keys or perform network calls.
Persistence & Privilege
Skill does not request always:true and does not modify other skills' configs. It writes/reads state under ~/.openclaw/workspace/memory which matches its function. The shell wrapper calls python in the bundle (i.e., spawns a process if you run it), but this is an optional convenience and is documented.
Assessment
This skill appears internally consistent for local token-optimization and model-routing. Before installing or running: 1) Review the bundle locally (it's small) — run the included sha256 integrity check (sha256sum -c .clawhubsafe) and inspect scripts. 2) If you don't want any subprocess invocation, avoid the optimize.sh wrapper and run the Python scripts directly. 3) Do not export provider API keys into your agent environment unless you intend the skill to detect and use those providers; providing keys enables network usage by other parts of your agent (the skill itself does not perform network calls according to its code). 4) Spot-check for network/system calls if you want extra assurance: grep for urllib|requests|socket|subprocess|os.system in scripts. 5) If you plan to apply config patches from assets/config-patches.json, review them before applying to openclaw.json. These steps will reduce risk while allowing you to benefit from the optimizer.Like a lobster shell, security has layers — review code before you run it.
auto-routevk978kj37vy3gcjq0qxnrhtnq0984a807latestvk978kj37vy3gcjq0qxnrhtnq0984a807model-routervk978kj37vy3gcjq0qxnrhtnq0984a807tool-detectionvk978kj37vy3gcjq0qxnrhtnq0984a807
License
MIT-0
Free to use, modify, and redistribute. No attribution required.