Claw Cost Router

Security checks across malware telemetry and agentic risk

Overview

This appears to be a purpose-aligned model-routing plugin, but it runs on startup, changes model choice for every turn, and locally logs prompt snippets by default.

Install only if you want a gateway-level plugin that automatically changes Claude model selection for every turn. Before enabling it, verify the plugin ID used for configuration, tune the skill lists and default model, and decide whether local prompt-snippet logging should be disabled.

VirusTotal

No VirusTotal findings

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

#
ASI02: Tool Misuse and Exploitation
Low
What this means

Misconfiguration or overly broad defaults could route ordinary chats or unknown skills to a cheaper or more expensive model than expected.

Why it was flagged

The plugin hooks model resolution and can override the model/provider for each turn. This is the advertised purpose, but users should understand it changes agent behavior globally while enabled.

Skill content
api.on("before_model_resolve", async (event, ctx) => { ... return { modelOverride: defaultModel, providerOverride: "anthropic" };
Recommendation

Review defaultModel, opusSkills, and haikuSkills before enabling, and test routing with non-sensitive prompts.

#
ASI06: Memory and Context Poisoning
Low
What this means

Sensitive information typed near the start of a message may remain in local OpenClaw plugin logs.

Why it was flagged

Each routing decision is appended to a local JSONL file, including channel/trigger metadata and the first 80 characters of the prompt.

Skill content
fs.appendFileSync(logPath, JSON.stringify(record) + "\n"); ... promptHead: prompt.slice(0,80)
Recommendation

Set logging to false if prompt snippets should not be retained, and periodically rotate or delete the decisions.jsonl file.

#
ASI09: Human-Agent Trust Exploitation
Info
What this means

A user may think they disabled logging or changed routing rules when the runtime plugin is still using defaults.

Why it was flagged

The documentation uses two different plugin entry names in the same configuration section, which may cause users to place settings under the wrong key.

Skill content
add a `config` block under `plugins.entries.claw-cost-router` ... "entries": { "model-router": {
Recommendation

Confirm the actual plugin ID with `openclaw plugins list` and verify the config key that OpenClaw applies.