Multi-Brain Protocol

Runs Kimi K2.5 and GPT 5.3 Codex in parallel pre-turn hook, injecting their perspectives for cognitive diversity before primary agent responds.

This package implements a daemon and hook that scan your OpenClaw session files, call external LLMs, and inject their output into agents' system/bootstrap context (and explicitly tells agents not to disclose those calls). Before installing, consider: 1) The registry metadata declares no required secrets, but the code reads local API key files (and may store keys plaintext) — audit and control where keys are stored. 2) The SKILL.md promises a keyword trigger, but the hook/daemon code runs on agent:bootstrap or continuously — it may run more often than documented. 3) Hardcoded paths (e.g., /Users/chadix/clawd/...) indicate the package was authored for a specific environment and may access unexpected files on your system. 4) The installer can register a persistent LaunchAgent/systemd service: only install if you trust the code and are comfortable giving it continuous read access to agent transcripts and local files. Recommendations: review the handler/daemon source carefully (search for any network calls and file reads); run it in a sandboxed account or VM first; remove or modify the 'never mention' instruction if you want transparency to users; require explicit, secure configuration for API keys (avoid plaintext defaults); and prefer local-only providers (Ollama) or explicit opt-in triggers. If you are unsure, do not install the service — test manually in foreground mode and inspect logs and created files first.

Type: OpenClaw Skill Name: openclaw-multi-brain Version: 1.0.0 The skill is classified as suspicious due to the implementation of persistent background execution via `launchd` (macOS) or `systemd` (Linux) services, as seen in `daemon/install.sh` and `src/cli.js`. Additionally, API keys are stored in plaintext in `~/.dual-brain/config.json`, which is a significant security vulnerability explicitly acknowledged in `BUILD-SUMMARY.md` and `CHANGELOG.md`. While the persistence is for the stated purpose of running a daemon, and the plaintext key is a vulnerability rather than an active attack, these are high-risk behaviors that warrant a 'suspicious' classification rather than 'benign'.