Ai Model Router
v2.0.0Automatically routes requests between two configured AI models based on task complexity, privacy needs, and user preferences for optimized AI usage.
⭐ 0· 210·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
Name/description (route between local and cloud models based on complexity/privacy) aligns with the code: detector reads local Ollama configs, router scores complexity and checks privacy, and context manager stores conversation state. No unrelated credentials or binaries are requested. Minor mismatch: SKILL.md and examples reference paths like skill/core/router.py and modules/..., but the repo files are skill/router.py, skill/detector.py, skill/context.py — code attempts to import modules.context / modules.detector which may raise ImportError (the router has a fallback). Also SKILL.md claims `core ~200 lines` while router.py is larger; these are inconsistency/accuracy issues but not inherently malicious.
Instruction Scope
Runtime instructions stay within the stated purpose (route tasks, detect privacy, optionally read local Ollama config and write ~/.model-router config/context). They do write/read files in the user's home (~/.model-router, ~/.ollama/models.json and /usr/share/ollama/models.json) and the context manager persists truncated message content. The SKILL.md examples reference paths that don't match the shipped files (may confuse automated installs or cause ImportError), and the SKILL.md instructs local CLI commands that will execute Python scripts on the user's machine — review those paths before running.
Install Mechanism
No install spec provided (instruction-only install), no external downloads, no brew/npm installs. The code is present in the package so installing will only place those files locally; there is no installer that fetches remote code.
Credentials
The package declares no required env vars, and most operations are local/read-only. However built-in fallback models indicate cloud models may require API keys (e.g., ANTHROPIC_API_KEY is referenced in a Model definition). The skill does not declare or request those env vars in metadata — using cloud models will require the user to supply credentials separately. Privacy detection will scan text for tokens/emails, which is expected behavior but could flag benign inputs.
Persistence & Privilege
The skill does create and write files under the user's home (~/.model-router/contexts.json and models.json) for configuration and conversation context — this is consistent with its purpose. It does not request global agent privileges, does not set always:true, and does not modify other skills or system-wide configuration in the visible code.
Assessment
This skill is largely coherent with its stated purpose, but review a few things before installing: 1) File/path mismatches: SKILL.md examples reference skill/core/router.py and modules/* while the files are under skill/*. Confirm how the package will be executed and fix paths before running. 2) The router will create ~/.model-router and write contexts and models.json — expect local persistence of truncated message text. 3) Cloud models in the built-in registry may require API keys (e.g., ANTHROPIC_API_KEY) even though the skill metadata doesn't declare required env vars; provide those only if you trust the cloud provider and the skill. 4) The privacy-detection regexes will flag emails, API keys, and tokens in user input and force routing to the primary (local) model — this is expected but may produce false positives. 5) Because some code was truncated in the provided bundle, if you plan to let the agent invoke this skill autonomously or route to cloud models, inspect the remaining code paths that actually perform network calls to model providers to ensure no unexpected endpoints or exfiltration are present. If unsure, run the skill in a sandboxed environment and inspect created files and network activity before adding it to agents that have access to sensitive data.Like a lobster shell, security has layers — review code before you run it.
latestvk97ep71y31hk3jg6tybpfnhkmx831bpd
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
