Skillv1.0.0

ClawScan security

Auto Model Router · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 5, 2026, 10:25 AM

Verdict: Benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's files and runtime instructions are consistent with its stated purpose (automatic model routing) and do not request unrelated credentials or install code, though it will save config under your OpenClaw skill directory and trigger an OpenClaw reload.
Guidance: This skill appears coherent and implements a model-routing/configuration helper. Before installing: (1) confirm you are comfortable the skill will write config files to ~/.openclaw/skills/auto-model-router/ and may trigger an OpenClaw reload; (2) understand that actual model API keys (OpenAI, Anthropic, etc.) are not provided by this skill and must be managed separately by your platform — ensure those credentials are stored securely elsewhere; (3) if you want to restrict the skill's ability to modify agent state, review OpenClaw's permission controls for skill file writes and reload actions. If any of the model names or default mappings are not appropriate for your environment, review and edit the included config/*.json and task-rules.json before enabling.

Review Dimensions

Purpose & Capability: okName, description, and included config files (plan-a/plan-b/default, task-rules) all align with a model-routing/configuration helper. There are no unexpected required binaries or environment variables that don't belong to a router skill.
Instruction Scope: noteSKILL.md describes classification rules, mapping logic, configuration commands, and a setup flow. It asks the agent to save configuration under ~/.openclaw/skills/auto-model-router/ and to 'auto reload OpenClaw' after setup — this is within scope for a skill that manages agent model assignments, but it does involve writing to the agent's skill directory and reloading the agent. There are no instructions to read unrelated system files, secrets, or to send data to third‑party endpoints.
Install Mechanism: okNo install spec or code files that would be downloaded/executed at install time; the skill is instruction-only. Lowest-risk install footprint.
Credentials: okThe skill declares no required environment variables or credentials. It references external model identifiers (openai/anthropic/etc.) but does not request API keys itself — management of those credentials appears to be left to the platform/user, which is proportionate for a routing helper.
Persistence & Privilege: okalways is false and disable-model-invocation is default (agent may invoke autonomously). The skill expects to write its own config files under ~/.openclaw/skills/auto-model-router/ (normal for a skill). It does not request system-wide privileges or claim to modify other skills' configs.