Smart Model Router

Route every substantive user request to the optimal LLM model. Trigger on EVERY substantive request. Score across 12 dimensions, pick the best model, switch...

Security Scan

The skill's code, runtime instructions, and requested filesystem access are coherent with its stated purpose (routing requests to the best model) and do not request unrelated credentials or installs.

Purpose & Capability

Name/description (model routing) match the included code and files: a 12-dimension scorer, a models database, and logic to read available models and pick one. Asking to read ~/.openclaw/openclaw.json to detect available models is proportionate to model-availability filtering.

Instruction Scope

SKILL.md only instructs running the local Python scorer, parsing its JSON output, optionally switching session model via session_status(), and to skip trivial messages. The metadata declares access to ~/.openclaw/openclaw.json, which the code actually reads. The instructions do not ask for unrelated files, credentials, or data exfiltration.

Install Mechanism

No install spec is provided (instruction-only). Code files are bundled with the skill; they are local and use only Python stdlib. No downloads, external URLs, or package installs are present.

Credentials

The skill declares no required environment variables or credentials. The code reads the user's home OpenClaw config (~/.openclaw/openclaw.json) and optional models/config JSON files — these are relevant to its job. No secrets or unrelated env access are requested.

Persistence & Privilege

Skill does not request always:true and does not modify other skills' configurations. It runs locally when invoked and relies on the agent to follow SKILL.md. The README suggestion to add a line to AGENTS.md to force compliance is advisory, not a capability change in the skill itself.

Assessment

This skill appears to do what it says: it scores messages and recommends a model, reading ~/.openclaw/openclaw.json to see which models are available. Before installing: - Run the provided --setup once and inspect the generated models.json and models database (capabilities_defaults.json) so automatic switching uses sensible defaults for your environment. If you don't configure models, the router intentionally returns fallback/default and will not switch. - Be aware the router executes on every substantive request (it will run the bundled Python script), which can add latency; test in a safe environment first. - The README suggests adding a note to AGENTS.md to reinforce agent compliance — avoid making the agent override other safety controls just to satisfy the skill. Only add such enforcement if you trust the skill and your agent's policy setup. - Confirm that ~/.openclaw/openclaw.json contains the model IDs you expect; the router filters out models not present there, which may cause no candidates if your config differs from the skill's database. No scan alerts were found in the supplied bundle, and no unrelated credentials or network endpoints are present in the code. If you want higher assurance, run the script locally and review its output on representative prompts before enabling automatic routing.

Every Substantive Request

1. Score: python3 <skill_path>/scripts/router.py --task "<user message>"
2. Parse JSON → get full_id, tier, confidence
3. If full_id is fallback/default: do NOT switch. Tell the user to run --setup.
4. If full_id differs from current model: session_status(model=<full_id>)
5. Answer the user. Do not volunteer routing details, but disclose when asked.

Skip for greetings, acknowledgments, heartbeats, or very short messages.

First-Time Setup

python3 <skill_path>/scripts/router.py --setup

Recommended: Reinforce in AGENTS.md

Add the following line to your agent's AGENTS.md for reliable auto-trigger:

Always follow smart-model-router SKILL.md instructions on every substantive request.

This is especially important when using less capable models that may otherwise ignore skill instructions.

CLI

--task TEXT    Route a task
--debug        Show scoring details
--setup        Generate models.json