Smart Model Selector
v1.0.0智能模型路由系统，根据任务自动选择最优 Qwen 模型（qwen3.5-plus/qwen-max/qwen-coder-plus），越用越聪明，节省成本
⭐ 0· 82·0 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 (automatic model routing among qwen variants) align with the code and SKILL.md. The hook inspects the first user message, chooses a model, records task stats, and exposes commands to view/reset/override — all expected for a model selector.
Instruction Scope
SKILL.md and hook instructions are scoped to analyzing the user message, suggesting/recording model choices, and replying to commands. The hook reads the first message and writes to a local data directory; it does not reference unrelated files, environment variables, or external endpoints in the visible code.
Install Mechanism
No install spec (instruction-only install via clawhub or manual copy). There is no network download/install step in the package itself. The README suggests git clone from a GitHub URL, which is normal for manual installs.
Credentials
The skill requires no environment variables, no credentials, and stores data in a local data/ directory. The DB path in the hook is within the skill workspace, matching the SKILL.md's local-storage claim.
Persistence & Privilege
always:false (normal). The skill is a hook that listens for session start and command events and will inject context at bootstrap when a first message exists. That means it can influence which model is used for a session, but it does not request elevated or persistent system privileges.
Assessment
This package appears to be a local-only model-routing helper that records task text and ratings into a local SQLite DB and injects model recommendations at session start. Before installing you should: 1) review the full src/model_selector.py (the provided file was truncated in the submission) to confirm there are no hidden network calls or data exfiltration paths; 2) confirm the repository/source URL and author identity if provenance matters; 3) be aware the skill stores task text locally (data/model_selection.db) — delete that file if you want to remove history; 4) note the hook injects context that may influence which model/endpoint the agent uses (ensure your agent/platform enforces model selection safely and that you understand any cost implications). If you need higher assurance, ask the author for the full source or an official repo link and re-run the review.Like a lobster shell, security has layers — review code before you run it.
latestvk972w9b4m9y48e7q95zpj7jrw183wh85
License
MIT-0
Free to use, modify, and redistribute. No attribution required.