Skip to content

GBrain Agent Operations — Setup Guide

Source: garrytan/gbrain (24,066 ⭐) Category: Agent Operations & Knowledge Management

GBrain is Garry Tan's opinionated agent brain layer for OpenClaw and Hermes Agent. This guide covers the operational skills that power agent identity, quality gates, scheduling, orchestration, and daily workflows.


Installation

Install the complete operational skillset in one shot:

# Core operational skills
npx skills add garrytan/gbrain/cross-modal-review
npx skills add garrytan/gbrain/soul-audit
npx skills add garrytan/gbrain/cron-scheduler
npx skills add garrytan/gbrain/daily-task-prep
npx skills add garrytan/gbrain/minion-orchestrator
npx skills add garrytan/gbrain/skillify
npx skills add garrytan/gbrain/skill-creator
npx skills add garrytan/gbrain/skillpack-check
npx skills add garrytan/gbrain/smoke-test
npx skills add garrytan/gbrain/frontmatter-guard
npx skills add garrytan/gbrain/repo-architecture
npx skills add garrytan/gbrain/signal-detector
npx skills add garrytan/gbrain/webhook-transforms

# Knowledge management
npx skills add garrytan/gbrain/perplexity-research
npx skills add garrytan/gbrain/article-enrichment
npx skills add garrytan/gbrain/concept-synthesis

Prerequisites

Requirement Details
GBrain installed git clone https://github.com/garrytan/gbrain.git ~/corpusiq-brain && cd ~/corpusiq-brain && bun install
Bun runtime GBrain requires Bun. Install: curl -fsSL https://bun.sh/install | bash
OpenAI-compatible endpoint For cross-modal-review: any OpenAI-compatible API (local Ollama, DeepSeek, Anthropic via proxy)
Hermes Agent v0.16.0+ recommended for full skill compatibility
Perplexity API key Required for perplexity-research skill only. Set PERPLEXITY_API_KEY in env.

Key Capabilities

Core Operational Skills

Capability How to Trigger Notes
Cross-Modal Review "second opinion", "cross-modal review", "double check this", "adversarial review" Spawns second model to review work before committing. Routes refusals silently.
Soul Audit "soul audit", "customize agent", "who am I", "set up identity" 6-phase interactive interview generating SOUL.md, USER.md, ACCESS_POLICY.md, HEARTBEAT.md
Cron Scheduler "schedule this", "set up a cron", "run daily at 6 PM" Manages staggered schedules, quiet hours, wake-up overrides
Daily Task Prep "morning prep", "what's on my calendar", "prepare for today" Calendar lookahead, meeting context loading, open threads synthesis
Minion Orchestrator "submit a gbrain job", "spawn agent", "check on agent", "what's running" Unified queue for shell jobs and LLM subagents. Durable, observable, steerable.
Skillify "turn this into a skill", "make a skill from this workflow" Meta skill: converts raw features into conformant SKILL.md files
Skill Creator "create a skill for X", "make a new skill" Template-driven SKILL.md generator with trigger mapping and tool declarations
Skillpack Check gbrain skillpack-check Generates agent-readable JSON health report on all installed skills
Smoke Test "run smoke tests", "verify environment" Post-restart validation with auto-fix for gbrain and OpenClaw
Frontmatter Guard Automatic on every brain write Validates and auto-repairs YAML frontmatter on brain pages

Knowledge Management Skills

Capability How to Trigger Notes
Repo Architecture "where should I file this", "which category" Decision protocol for filing brain pages by domain and lifecycle
Signal Detector Always-on, fires on every inbound message Detects actionable signals from messages, emails, notifications
Webhook Transforms "set up webhook for X", "ingest from webhook" Converts external events (SMS, meetings, social) into structured brain pages
Perplexity Research "research X with brain context", "what's new about Y" Brain-augmented web research — returns what's NEW vs what the brain already knows
Article Enrichment "enrich this article", "structure this dump" Transforms raw text dumps into structured pages with executive summaries
Concept Synthesis "synthesize concepts", "map my ideas" Deduplicates and organizes concepts into tiered intellectual maps

CorpusIQ Use Cases

Use Case How
Agent identity governance Run soul-audit to generate/update agent SOUL.md after any behavior change. Frontmatter-guard prevents metadata corruption across 400+ brain pages.
Pre-deployment quality gates Cross-modal review on all public content and code before pushing. Smoke-test after every Hermes restart to verify toolchain health.
Scheduled content operations Cron-scheduler manages 6 PM daily reports, morning video generation, and weekly performance audits with quiet-hour awareness.
Multi-agent fan-out Minion-orchestrator spawns parallel research subagents for competitive analysis, prospect research, and content mining — all from a single queue.
Skill development pipeline Skillify converts proven workflows into reusable skills. Skill-creator generates conformant SKILL.md. Skillpack-check audits the installed skillset.
Daily intelligence brief Daily-task-prep loads calendar + open threads + brain context. Signal-detector surfaces high-priority inbound signals. Combined, this replaces manual morning triage.
Knowledge synthesis Concept-synthesis maps ideas across brain sections. Article-enrichment structures raw research. Together, they turn information firehose into actionable intelligence.

Troubleshooting

Issue Fix
cross-modal-review: "no second model configured" Set OPENAI_BASE_URL and OPENAI_API_KEY in your environment. The skill spawns any OpenAI-compatible model.
soul-audit: generates empty files Ensure the templates/ directory exists at SKILL_DIR/skills/soul-audit/templates/. The skill reads template scaffolds, not ships pre-filled content.
minion-orchestrator: jobs stuck in queue Run gbrain jobs list to see active jobs. Use gbrain jobs cancel <id> for stuck jobs. Restart the gbrain daemon if queue is unresponsive.
skillpack-check: returns empty report Run gbrain skillpack-check --rebuild-cache to regenerate the skill index. First run after install may need cache priming.
"gbrain: command not found" after install Add export PATH="$HOME/.bun/bin:$HOME/corpusiq-brain/bin:$PATH" to your shell profile. The CLI lives in the repo's bin/ directory.
frontmatter-guard: false positive on valid YAML The validator enforces strict GBrain schema. Custom frontmatter fields outside the schema trigger warnings. Run gbrain schema-author --add-field <name> to register new fields.

Verification

# Verify GBrain is installed and reachable
cd ~/corpusiq-brain && gbrain --version

# List all installed gbrain skills
gbrain skillpack-check 2>&1 | head -20

# Test cross-modal review (will use current model if no second configured)
echo "test" | gbrain cross-modal-review --dry-run

# Run smoke tests
gbrain smoke-test

Skills Catalog | Discovery Page Curated by CorpusIQ — one MCP endpoint, all your business tools. Content remains attributed to original authors and repositories.