Skip to content

HANDOFF.md

Latest commit

 

History

History
171 lines (125 loc) · 9.65 KB

HANDOFF.md

File metadata and controls

171 lines (125 loc) · 9.65 KB
Raw

connor-stack — Agent Handoff

You're a fresh Claude Code session inside E:\Dev2\connor-stack. This file gets you fully oriented. Read it top to bottom before touching anything else.

What this repo is

A public, choose-your-own-adventure portal at cgallic/connor-stack (not yet pushed to GitHub) that documents how Connor Gallic uses AI — through his actual proofs of concept, not as prescriptive playbook.

It's a hub. Working subsystems (brain, news digest, marketing skills, shorts pipeline, etc.) live in their own existing public repos. This repo connects them with a guided experience the visitor explores, then forks pieces of into their own setup.

Three audiences:

  • A — Solo marketers / agency owners copy client folder patterns, content workflows
  • B — Claude Code power users / agent builders copy the orchestration patterns, brain loop, skills/hooks
  • C — Curious readers take a tour of how it all fits

Core design philosophy: "Their Claude finishes the setup." We ship structured prompts + concept demos. The visitor pastes one into their own Claude and it drives configuration locally. We provide skeleton + inspiration, not a fat installer.

Where the work currently stands

Done

  • Repo skeleton at E:\Dev2\connor-stack (git init, MIT LICENSE, .gitignore, one bootstrap commit on main)
  • Spec doc: E:\Dev2\CMO_Agent_System\docs\superpowers\specs\2026-05-22-connor-stack-design.md (authoritative; reflects all decisions including spoke list and out-of-scope items)
  • Voice anchor — three markdown files written by a subagent earlier in this conversation:
    • README.md (581 words) — three-door GitHub landing
    • docs/philosophy.md (1,311 words) — "How Connor Does AI" essay
    • docs/tour.md (688 words) — architecture tour with ASCII hub+spokes diagram

These three files have NOT been reviewed by Connor yet. Read them first to understand voice, but assume they may need rework once we know the new shape.

Pivoted (mid-conversation, before any HTML work)

The repo is no longer a markdown docs collection. It's a choose-your-own-adventure HTML experience hosted at a URL (probably GitHub Pages, possibly a custom domain).

Why: Connor wants the visitor to feel like they're exploring proofs of concept that "wrap them in a giant hug with AI" — interactive, personal, not prescriptive. Markdown docs read as "Connor is telling you the One True Way." HTML CYO reads as "Connor is showing you what he built; remix it."

The voice anchor markdown still has value:

  • philosophy.md → likely repurposes as the long-form "Connor's actual thinking" page behind a CYO branch
  • README.md → reframes to a GitHub-renders fallback that points visitors to the live site
  • tour.md → likely becomes one of the HTML pages

Cut from v1

  • ralph-wiggum-marketer plugin (deprecated, per Connor)
  • KaiCalls / Case Engine internals (proprietary)
  • Pendant / Jarvis layer (too fresh, too Connor-specific)
  • Fat installers / one-command runtime (out of scope — visitor's Claude finishes setup)

The choose-your-own-adventure shape (rough)

index.html  [interview video embed + intro]
  "How I built an AI that fits MY life.
   Here are the pieces. Make them fit yours."

  → I want to ship better client work       (Marketer path)
  → I want to build the machine that         (Agent-builder path)
    builds the machine
  → I'm just curious what someone's          (Tour path)
    actually running

Each concept page is its own HTML artifact:
  - Concept name + the "why" Connor built it
  - Working demo / screenshot / interactive piece
  - Actual code or config snippet (copy button)
  - "Make it yours" — fork the spoke, paste this prompt into your Claude
  - Cross-links to siblings (CYO branching, not linear)

Each concept page ships with an embedded markdown prompt block ("paste this into your Claude") that drives a personalized version locally. No chat widget on the site; the visitor's own Claude does the personalization.

Spokes — the components this repo points at

Already-public (link, don't vendor)

Repo One-line
cgallic/open-brane events.db brain foundation, MCP for agents
cgallic/paperboy news/papers Discord digest, daily
cgallic/kai-cmo-harness open-source AI CMO for Claude Code (marketing skills)
cgallic/blackbox self-improving feedback loop for Claude Code
cgallic/diwan preflight tests / decision receipts
cgallic/zehrava-gate safe commit layer for AI agents
cgallic/cartoonimator deterministic mascot lipsync
cgallic/mcp-browser terminal-based MCP installer
cgallic/shorts-pipeline daily multi-channel YouTube Shorts pipeline
cgallic/mydeadinternet, dead-internet-core, mdi-mcp-server, mdi-oracle-solana MDI ecosystem
cgallic/awesome-meetkai curated OpenClaw deployments
cgallic/avp Agent Verification Protocol
cgallic/wake-up-skill ClawdHub skill

To create (only one remaining)

  • cgallic/agent-webhook-gateway — extracted from E:\Dev2\CMO_Agent_System\gateway\ (FastAPI service exposing the Python automation scripts as HTTP endpoints)

Authoritative references

Read these when you need source material or to verify decisions:

Source What's there
E:\Dev2\connor-stack\INVENTORY.md COMPREHENSIVE MAP of every system, repo, service, concept in the stack — read this second after HANDOFF. Includes OpenClaw collective, Hermes architecture, brain stack, pendant layer, A2A peers, frameworks.
E:\Dev2\CMO_Agent_System\docs\superpowers\specs\2026-05-22-connor-stack-design.md The locked spec — purpose, audience, architecture, success criteria
E:\Dev2\CMO_Agent_System\CLAUDE.md Frameworks, personas, channels, checklists, knowledge-base structure (source material for "marketer door" concept pages)
C:\Users\cgall\.claude\projects\E--Dev2-CMO-Agent-System\memory\ Connor's persistent memory — preferences, project state, infra map, voice rules
E:\Dev2\connor-stack\docs\philosophy.md The voice-anchor essay written this session (use as voice reference)
https://github.com/cgallic/paperboy Voice reference — one-line README hook style
https://github.com/cgallic/open-brane Voice reference — punchy, concrete
https://github.com/cgallic/shorts-pipeline Voice reference — recently shipped, full layered README

Connor's voice — non-negotiables

From memory + observed patterns. Do not violate.

  • No emojis in any repo file
  • No AI-tell phrases: "in conclusion", "it's worth noting", "leverage", "robust", "seamless", "ecosystem", "fragmentation"
  • No "It's not X. It's Y." constructions (Connor banned this in his short-form scripts)
  • No rhetorical setup-payoff arcs or three-beat rhythms — they read as AI
  • No bulleted "Why this matters" sections that read as marketing copy
  • Punchy one-line pitches in the open-brane / paperboy style — contrast old-thing / new-thing or pain / fix
  • First person is fine in essays where it earns trust (philosophy doc)
  • Concrete nouns over abstractions: events.db, ~/.claude/projects/, ffmpeg, Rhubarb — name real tools
  • Short sentences, fragments OK, mix into short paragraphs
  • ASCII diagrams in fenced code blocks, not Mermaid

Open questions Connor hasn't answered yet

When you have a coherent design proposal, raise these with him before building further:

  1. Hosting — GitHub Pages on cgallic.github.io/connor-stack is the zero-effort default. Or a custom domain (connorstack.com, connor.codes, something else)?
  2. HTML aesthetic — pull from the frontend-design skill (distinctive, production-grade, anti-generic-AI), or match a specific reference site?
  3. Interview video — does it get recorded before launch (blocks ship) or after (placeholder for now)?
  4. License on new spokes — assume MIT to match existing repos unless Connor says otherwise.

Suggested next steps (in order)

  1. Read INVENTORY.md — comprehensive system map. Without this you'll under-scope.
  2. Read the voice anchor — open README.md, docs/philosophy.md, docs/tour.md. Get the voice in your head.
  3. Get Connor's call on the open questions above — at minimum hosting + aesthetic, since they shape the build.
  4. Sketch the CYO map — using INVENTORY as input, list every HTML page that needs to exist, branching structure, what each demonstrates. One markdown table is enough. Must cover the OpenClaw collective and the Hermes architecture, not just brain + marketing.
  5. Build the entry page (index.html) as a single-page proof of concept first. Get Connor's reaction before generating the rest.
  6. Fan out concept pages in parallel once entry page is approved — subagent per page, each agent reads HANDOFF + INVENTORY + the spec + the approved entry page as voice/structure anchor.
  7. Extract agent-webhook-gateway into E:\Dev2\agent-webhook-gateway\ as a side task while content is being built.
  8. Write setup-prompts/ — the markdown prompts visitors paste into their own Claude. These are the "wrap them in a giant hug" pieces.
  9. Write links.md last, once every spoke has been confirmed real and reachable.
  10. gh repo create cgallic/connor-stack --public + push when content is solid.

What this repo is NOT

  • A turnkey product ("clone, run, done")
  • A prescriptive "the right way to do AI" playbook
  • A monorepo containing all of Connor's code
  • A managed service
  • A Discord community / paid course / support tier

It is: a curated portal of proofs of concept that demonstrate how one person uses AI in their life, designed for others to remix.