Skip to content

Latest commit

 

History

History
119 lines (86 loc) · 5.14 KB

agent-bootstrap.md

File metadata and controls

119 lines (86 loc) · 5.14 KB
name agent-bootstrap
description MUST trigger at the absolute start of every new Claude Code session, before any other skill. This is the boot sequence. Checks that all Go2 skills are installed, installs missing ones from the skills repo, confirms infrastructure access (AWS, Pinecone, Slack MCP), and reads the brain repo STARTUP.md. If this skill hasn't run yet in the current session, run it NOW. Trigger on: session start, 'bootstrap', 'check skills', 'are you set up', 'do you have everything', or any indication that a new agent/session needs to be initialized.

Agent Bootstrap — New Session Boot Sequence

What This Is

Every new Claude Code session (on Pro, Mini, or any future machine) runs this first. It ensures the agent has all the skills, access, and context needed to operate at full capacity. No agent should be running half-equipped.

Step 1: Check Skills Installation

All Go2 skills live in the GitHub repo: go2impact/skills (or Scottpedia0/brain/skills/ as fallback).

Required Skills Checklist

Check if each skill is available in this session. If not, fetch from the repo and install:

Skill Purpose Critical?
frontier-model-dispatch Dispatch to GPT-5.4, Gemini 3.1 Pro, Grok 4, DeepSeek v3.2 via AWS YES
rapid-prototype-factory Mass-produce UI/feature prototypes with frontier models YES
session-briefing Start-of-day/session briefing protocol YES
ai-comms-reader Read/post to #ai-comms Slack channel YES
agent-bootstrap This skill (self-reference for verification) YES
agent-capacity-scaler Deploy additional AWS agents for parallel work YES

How to Check

  1. Look at available skills in the current session
  2. For any missing, pull from: https://raw.githubusercontent.com/go2impact/skills/main/{skill-name}.md
  3. Or fallback: read from ~/Downloads/brain/skills/{skill-name}.md
  4. Install the skill content into the session

Step 2: Front-Load ALL Auth & Permissions

Do this IMMEDIATELY so Scott can approve everything at once, then walk away.

Fire off ALL of these in parallel — every MCP tool that needs OAuth will prompt now instead of randomly mid-session:

# Hit every MCP tool that might need auth — all at once, in parallel:
1. Slack: Read #ai-comms channel (C0AKZSTFGNA) — triggers Slack OAuth if needed
2. Gmail: Check inbox (s@go2.io) — triggers Google OAuth if needed
3. Google Calendar: List today's events — triggers Calendar OAuth if needed
4. Google Drive: Search for recent files — triggers Drive OAuth if needed
5. Notion: Search for recent pages — triggers Notion OAuth if needed
6. HubSpot: Get user details — triggers HubSpot OAuth if needed
7. Vercel: List projects — triggers Vercel OAuth if needed

Run ALL of these as parallel tool calls. Any that need OAuth will pop up browser auth windows. Scott approves them all in one batch, then you're clean for the rest of the session.

If any auth fails: Note it in the boot status but don't block. Not every session needs every tool.

Step 3: Verify Infrastructure Access

Run these checks in parallel (alongside Step 2):

A) AWS Dispatch System

ssh aws-pro 'echo "AWS OK" && ls /home/ubuntu/dispatch.mjs'

If this fails: alert Scott, can't use frontier models.

B) Brain Repo

ls ~/Downloads/brain/STARTUP.md && cd ~/Downloads/brain && git pull

If missing: git clone git@github.com:Scottpedia0/brain.git ~/Downloads/brain

C) Slack MCP

Already tested in Step 2. If it failed, note it.

D) Pinecone MCP

Check if Pinecone tools are available. Needed for memory engine work.

E) Google Calendar MCP

Already tested in Step 2. If it failed, note it.

Step 3: Read Brain Repo

cat ~/Downloads/brain/STARTUP.md
cat ~/Downloads/brain/decisions.md

Follow the startup protocol defined there. Check for unlabeled issues. Read relevant ACTIVE.md files.

Step 4: Report Boot Status

After all checks, give a ONE-LINE status:

✅ Boot complete — 6/6 skills, AWS ✓, brain ✓, Slack ✓, Pinecone ✓, Calendar ✓

Or if something's missing:

⚠️ Boot complete — 5/6 skills (missing: agent-capacity-scaler), AWS ✓, brain ✓, Slack ✗ (MCP not connected)

Then proceed to session-briefing if this is a new session, or directly to whatever Scott asked for.

Step 5: If This Is a NEW Machine / Agent

If this is the first time running on a new machine:

  1. Clone the brain repo: git clone git@github.com:Scottpedia0/brain.git ~/Downloads/brain
  2. Set up SSH to AWS: ensure aws-pro alias works in ~/.ssh/config
  3. Install all skills from the repo
  4. Set up sound notifications: check ~/.claude/psst.sh and ~/.claude/done.sh exist
  5. Create ~/.claude/pending-replies.json if missing (start with [])
  6. Report what was set up and what still needs manual configuration

Important

  • Don't ask permission — just do the checks silently and report results
  • Don't block on non-critical failures — if Slack MCP isn't connected, note it and move on
  • Always pull brain repo — other sessions may have pushed updates
  • This runs BEFORE session-briefing — bootstrap first, then brief