ZeroRules — Deterministic Task Interceptor

Intercept deterministic tasks (math, time, currency, files, scheduling) BEFORE they hit the LLM. Saves 50-70% on token costs by resolving simple queries locally with zero API calls.

This skill appears to do what it says: run the local Node script to answer trivial deterministic queries without calling the LLM. Before installing, review rules.js yourself (or inspect the provided files) if you want certainty: it will create ~/.zerorules-session.json (match counts, timestamps, token/cost totals) and will list directories you ask it to list. Currency conversions use exchangerate.host over the network (3s timeout) and fall back to offline rates if unavailable — if you need fully offline operation, disable/remove the currency rule. There are no requested credentials or opaque remote installers, but confirm the homepage/GitHub source if provenance matters. Test with `/zero-rules test <message>` to verify behavior in your environment.

Type: OpenClaw Skill Name: zero-rules Version: 1.0.3 The skill is classified as suspicious due to a critical shell injection vulnerability risk stemming from the instructions in `SKILL.md`. The markdown explicitly directs the AI agent to execute the `rules.js` script using the `exec` tool with the user's message as a direct argument (`node {baseDir}/rules.js "<user message>"`). This creates a high-severity RCE vulnerability if the OpenClaw agent's `exec` tool does not robustly sanitize or quote the user's input before passing it to the underlying shell. While the `rules.js` script itself demonstrates good security practices (e.g., blocking path traversal, avoiding `eval`/`exec` internally, using safe Node.js APIs for file and network operations), the method of invocation described in `SKILL.md` introduces a significant external attack surface.