gbrain

name	gbrain
description	GBrain CENTRAL ORCHESTRATOR — 19-layer health dashboard (incl. AWS infra) + 7th-phase compounding engine + upstream features watch + skill propagation. THIS skill is source of truth for the GBrain stack: when upstream gbrain/openclaw releases new features (voice, zoom, dream synthesize, etc.), Layer 17 surfaces them as informational and tells which satellite skills (brain-write-macro, signal-detector, gbrain-http-wrapper) are affected. Subcomandos: /gbrain check (default), /gbrain fix, /gbrain news, /gbrain bugs, /gbrain compare, /gbrain save, /gbrain compound, /gbrain bootstrap, /gbrain principles, /gbrain manifest, /gbrain manual, /gbrain custom-instructions. Triggers: /gbrain, revisa gbrain, salud del brain, verifica el brain, gbrain status.
allowed-tools	Bash Read Write
user-invocable	true
distribute-to	["claude","openclaw"]
role	central-orchestrator
related-skills	["brain-write-macro","signal-detector","gbrain-http-wrapper"]

/gbrain — GBrain Central Orchestrator

You are running the central authority for Sergio's GBrain + OpenClaw stack. This skill does NOT just report — it dictates how related skills behave.

Central orchestrator contract (READ FIRST)

When upstream gbrain or openclaw changes anything, propagation flows from this skill outward:

1. gbrain CLI upgrades → Layer 17 surfaces new features (read from upstream commits) and tells the user which satellite skills are affected. Format: you have a module for X — try it with <command> OR not applicable to your setup.
2. Schema migrations → Layer 3b detects DB column drift and flags wrapper/skills that read affected columns.
3. Wrapper changes (gbrain-http-wrapper) → MANIFEST.json tracks expected version + security features; Layer 11b verifies wrapper health + public Tailscale Funnel TLS; bootstrap.sh confirms canonical install.
4. Sub-skill updates (brain-write-macro, signal-detector) → Layer 5 confirms loaded; Layer 11b D verifies macro pointer in SOUL.md; if upstream gbrain ships native equivalent, Layer 17 flags "subskill no longer needed".
5. OpenClaw new features (voice, zoom, video, etc.) → Layer 17 reads upstream commits/tweets, surfaces as informational, with concrete try-it commands. NOT urgent unless user-actionable.
6. Compound engine → owns subcommand /gbrain compound, runs as 30-min-after-dream complement. Since gbrain v0.23.0 ships synthesize+patterns natively, compound is now POST-DREAM (DB-side analysis), not the only LLM-driven layer.

Hard rule: if a satellite skill needs to change because of upstream movement, this skill MUST flag it first via Layer 17b (skill propagation). Never modify brain-write-macro/SKILL.md or signal-detector/SKILL.md without checking that /gbrain check says "skill propagation: needed".

Action

Execute the runnable script that already exists at ~/.openclaw/skills/gbrain/run.sh. It supports subcommands:

bash ~/.openclaw/skills/gbrain/run.sh check                # full 19-layer dashboard + alert banner
bash ~/.openclaw/skills/gbrain/run.sh verify               # lie-detector: re-check each claim vs ground truth
bash ~/.openclaw/skills/gbrain/run.sh fix                  # auto-fix safe issues (idempotent)
bash ~/.openclaw/skills/gbrain/run.sh news                 # only upstream releases/PRs/issues
bash ~/.openclaw/skills/gbrain/run.sh bugs                 # only bugs that affect THIS user
bash ~/.openclaw/skills/gbrain/run.sh compare              # diff vs previous snapshot
bash ~/.openclaw/skills/gbrain/run.sh save                 # full + save to ~/brain/reports/gbrain-YYYY-MM-DD.md
bash ~/.openclaw/skills/gbrain/run.sh bootstrap            # verify entire stack canonical
bash ~/.openclaw/skills/gbrain/run.sh principles           # operational rules + decision log
bash ~/.openclaw/skills/gbrain/run.sh manifest             # canonical inventory
bash ~/.openclaw/skills/gbrain/run.sh manual               # full manual with use-cases
bash ~/.openclaw/skills/gbrain/run.sh custom-instructions  # generate Custom Instructions block

bash ~/.openclaw/skills/gbrain/run.sh compound run         # run compounding cycle
bash ~/.openclaw/skills/gbrain/run.sh compound dry-run     # preview proposals
bash ~/.openclaw/skills/gbrain/run.sh compound status      # confidence per category
bash ~/.openclaw/skills/gbrain/run.sh compound history     # last 10 cycles
bash ~/.openclaw/skills/gbrain/run.sh compound revert <id> # queue revert

How to invoke

1. If user just types /gbrain → run bash ~/.openclaw/skills/gbrain/run.sh check and stream output as markdown directly.
2. If user types /gbrain <subcommand> (e.g. /gbrain fix, /gbrain save, /gbrain compound status) → pass that subcommand verbatim.
3. If user says natural language ("revisa gbrain", "salud del brain", "está bien gbrain?") → run check by default.

Output handling

The script outputs Markdown directly. Stream that output as-is — no reformat unless asked. If the script prints errors:

• gbrain: command not found → bun install -g gbrain@github:garrytan/gbrain --force
• openclaw gateway status returns "not running" → openclaw gateway start
• psql connection fails → check ~/.gbrain/config.json has valid database_url

19-layer surface (what each layer does)

• 🚨 Alert banner (top, only when critical): stuck sessions >0, openclaw-node sin API keys, model sin fallbacks, queue depth >100, public Tailscale Funnel TLS desync, AWS infra drift
• L1 Versiones (gbrain + openclaw vs latest)
• L2 Runtime (gateway, telegram conns, npm loops, modelo primario + fallbacks, MCP, SOUL)
• L3 Doctor (gbrain doctor structured)
• L3b Schema correlation (DB ↔ wrapper ↔ skills, catches silent migration drift)
• L4 Stats (pages/chunks/links/timeline)
• L5 Skills loaded
• L6 Captura ambient 24h (signal-detector effectiveness)
• L7 Bugs upstream que TE afectan
• L8 News (releases/PRs/issues with date + labels)
• L9 Snapshot diff
• L10 Canonical files mtime
• L11 MCP Health
• L11b Wrappers + Integrations (HTTP wrappers, recipes, OAuth clients, capture by source)
• L12 Stuck sessions detector
• L13 Process env audit
• L14 Cron failure rate 24h
• L15 Upstream changelog
• L16 Upgrade Decision Engine (INSTALAR/ESPERAR/SKIP per tool)
• L17 Upstream Features Watch — tweet-style feature surface + skill propagation rules
• L17b Skill propagation — concrete propagation hints per satellite skill
• L18 Clientes MCP gbrain + custom-instructions adoption
• L18b Modelo activo cross-runtime (live read)
• L19 AWS Infra Health (NEW) — 9 controles (CloudTrail, GuardDuty, Budget, DLM, SG, EBS, IMDSv2, root keys, GBrain stack/jarvis-v3 page) + alert si drift

Files

• SKILL.md — this file (central orchestrator descriptor)
• run.sh — single source of truth, all subcommands route here
• bootstrap.sh — idempotent stack verifier
• MANIFEST.json — canonical inventory (versions, paths, schema contracts, upstream issues)
• MANUAL.md — full manual with use-cases
• PRINCIPLES.md — operational rules + decision log
• compound/run.sh + compound/morning-report.sh + compound/prompts/analyze.md — 7th-phase compound engine
• tools/openclaw-to-corpus.py — converter for ~/.openclaw/agents/*/sessions/*.jsonl → .txt corpus that gbrain v0.23.0 dream synthesize reads

Reference docs

• PRINCIPLES.md — operational rules (canonical-wins, no-signal-masking, decision log with dates)
• MANIFEST.json — schemas, expected versions, security gaps, upstream issues open
• Snapshots: ~/.openclaw/skills/gbrain/snapshots/ (auto-generated)

Note on duplication (Claude Code vs OpenClaw)

This file lives at ~/.openclaw/skills/gbrain/SKILL.md and is mirrored to ~/.claude/skills/gbrain/ via the monorepo durang/skills. Both invocation paths (Claude Code /gbrain AND OpenClaw Telegram /gbrain) call the same run.sh. Single source of truth: run.sh.