brain-write-macrolisted

Explicit save macro — when user says 'guarda en gbrain' (and 23 phrase variants in ES/EN), automatically extract entities, decisions, originals, and links from the conversation and write them as gbrain pages. Companion to signal-detector (which captures passively per-message); brain-write-macro is the explicit user-triggered version that works in clients without hooks (Claude Desktop, claude.ai web, mobile).
durang/skills · ★ 1 · AI & Automation · score 70

Install: claude install-skill durang/skills

# /brain-write-macro — Explicit save by phrase (works on every client) ## Why this skill exists **The capture gap:** GBrain has two automatic capture paths today: 1. **Claude Code CLI** → Stop hook (`~/.claude/settings.json` → `signal-detector.py`) fires at end of every turn, runs Haiku over the transcript, writes pages. 2. **Telegram (Harviz/OpenClaw)** → SOUL.md instructions tell the model to extract entities and call `gbrain__put_page` on every inbound message. **But Claude Desktop, Claude.ai web, ChatGPT, and Codex CLI have NO automatic hooks.** The model only writes to gbrain if: - The user pastes Custom Instructions telling it to, OR - The user explicitly says: "guarda en gbrain" This skill defines the **canonical behavior** for the explicit-phrase trigger so all clients implement it identically. ### Client delivery map (where instructions live per client) | Client | How rules are delivered | Format | |---|---|---| | **Claude Code CLI** | Stop hook (signal-detector.py) — automatic | Full v3 (embedded in Haiku prompt) | | **claude.ai web/app** | Settings → Profile → Custom Instructions | Full v3 (via `/gbrain custom-instructions`) | | **ChatGPT App** | Settings → Personalización → Custom Instructions | Compact v3 (~780 chars, trigger: "guarda en gbrain") | | **Codex CLI** | `~/AGENTS.md` (gbrain section) | Full v3 (read automatically) | | **OpenClaw / Telegram** | SOUL.md + this skill loaded as skill | Full v3 (model reads SKILL.md) | | **Hermes** | Inherited via