The operating system for founders who run more than one thing.
Claude Code runs your portfolio like a CTO would: plans before it spends, asks before it ships,
remembers every project, and puts every decision that needs you one tap away on your phone.
You have a product that pays the bills, a side venture, a partner project and a day job. Claude Code is brilliant at any one of them for an afternoon. Across all of them, for months, it forgets what it was doing, spends what it should not, and asks you the same questions again. Mission Control is the missing layer: a portfolio memory, a set of rules it cannot argue its way around, and a control room you can run from your phone.
The Mission Control App on a phone: Flight's morning summary and the Controller running each front.
Your main Claude Code session becomes Flight, and it staffs a Controller for each venture: a long-lived agent that works that one front, keeps its own back room of helpers within a budget you set, and reports in four lines (state, shipped, blocked on you, next), plus a fifth when something other fronts depend on has changed. Change a Controller's model mid-job, call it, or stand it down with a written handoff. PR watches wake a Controller when its checks or reviews change, so nobody polls GitHub.
When a Controller needs you to decide something, Flight asks you as its own question, with the Controller's options spelled out, each with what it does; urgent ones first. Pick one, or answer in your own words, and the pick goes straight to that Controller. Things only you can do land in Flight as to-dos with their links, and Done tells the Controller. Changes that ripple into other fronts are routed to the Controllers they affect. Nothing waits silently.
/console app starts the Mission Control App: Flight's conversation with its questions and
your to-dos, your Controllers and the message traffic, as a home-screen app. Over Tailscale it reaches your phone
in any city, signed in by your own Tailscale account and never exposed to the public internet.
Send Flight a message or a photo, answer the questions it asks, get an alert when a decision
lands. One front door covers every Claude account you use (work, personal, a partner's) with
a switcher in the header. Once on, it keeps itself running for as long as your session does.
It never commits, pushes, deploys or spends without your word in the moment. Fan-outs of agents are sized and approved before they fire, every agent gets the cheapest model and effort that holds the bar, and evidence comes before every claim. Controllers may push their own feature branches and open PRs, never merge, never touch a base or deploy branch, and a guard enforces that on every tool call. Hooks block personal data and process chatter from ever reaching a commit or a PR.
One ticket, one epic or a whole product, each with the same standards inherited all the way down:
flowchart LR
P["/product-flow<br/>idea to launch<br/>five gates you approve"] --> E["/epic-flow<br/>a locked plan of tickets<br/>one PR for the whole"]
E --> M["/mission-flow<br/>one ticket to a green PR<br/>Phase 0 to 8"]
M --> PR(["a reviewed, tested PR<br/>waiting for your merge"])
/mission-flow takes a bug or a task to a checks-green PR: root cause before fix, tests at
the altitude of the risk, review by right-sized agents, the real app run before anything is
called done. /epic-flow plans work bigger than one ticket into a tree you lock, then runs
each piece as a full mission-flow. /product-flow takes an idea through framing, research,
design, architecture, build, deploy, beta and launch, and stops at the five moments where a
human has to say yes.
Every front gets a hub, every repo a spoke with its gotchas, every day a single pointer to what matters. Resume any project with no re-briefing. Each session's lessons pass an eight-rule critique into a learning log, and the ones that keep proving true are promoted, with your sign-off, into standing rules that load every session. It ships with a seed corpus of 23 lessons paid for on a real multi-venture portfolio.
flowchart TB
YOU(["You<br/>terminal or phone"]) <-->|messages, answers, photos| FLIGHT["Flight<br/>your main Claude Code session"]
FLIGHT -->|"/console staff"| C1["Controller<br/>storefront"]
FLIGHT --> C2["Controller<br/>billing-api"]
FLIGHT --> C3["Controller<br/>mobile-app"]
C1 & C2 & C3 -->|status, questions, to-dos| FLIGHT
C2 --> BR["back room<br/>helper agents within budget"]
subgraph PORTFOLIO["Your portfolio, plain Markdown in git"]
HUB["hubs, spokes, boards,<br/>today's pointer, the rules"]
end
FLIGHT --- PORTFOLIO
Everything is local and yours: plain Markdown files in a git repository, the rules Claude loads every session, and 26 skills that turn your rituals into one command each. No hosted service, no database, no lock-in.
git clone https://github.com/<owner>/mission-control my-portfolio
cd my-portfolio && claude
> hiThat is the whole install. Your first message starts LIFTOFF, a guided session (about
4 to 6 minutes per project, three approval gates) that interviews you, scans your real repos
instead of quizzing you about them, and stands up your portfolio. It ends with a proof: a fresh
session reads your setup cold, states your board and your rules, and refuses an
unauthorized push. That refusal is the product working. Big portfolio? Lift off with the
fronts you touch this week and add the rest later with /new-front.
Then, for the phone: load the mission-fleet plugin (see tools/README.md),
type /console app, and follow the remote access guide.
| When | You type | What happens |
|---|---|---|
| Morning | /briefing |
It reads every front and locks today's one objective |
| Any time | /console staff <front> |
A Controller takes a venture and runs it |
| A bug or a task | /mission-flow <description> |
Root cause to a checks-green PR, nothing merged without you |
| Bigger work | /epic-flow or /product-flow |
A plan you lock, then the whole thing delivered |
| Before a meeting | /standup |
A paste-ready report with a two-minute talk track |
| Evening | /debrief |
Progress written, lessons captured, tomorrow's pointer set |
All 26 skills, grouped by what they are for: .claude/skills/README.md.
- Not an autonomous coder you leave alone. It presents, you decide. Every irreversible or outward-facing step stops for you, from your phone if you are away.
- Not a hosted product. It runs in your Claude Code, on your machine, in your files.
- Not a prompt pack. The rules are enforced by hooks and guards, not just written down.
Keep this repository as upstream and your portfolio on a private origin:
git remote rename origin upstream
git remote add origin <your-private-repo-url>
git pull upstream main # updates framework/, .claude/skills/ and tools/; your _command/ never collidesRead the new entries in CHANGELOG.md after each pull: doctrine changes never
reach you silently. Your operating instructions live in _command/CLAUDE.operating.md, which
the shipped CLAUDE.md imports, so you never edit CLAUDE.md and no pull meets a local edit.
An instance lifted off before 2026-10-17 kept them in CLAUDE.md itself: move them once, before
your next pull, with the steps in the CHANGELOG entry of that date. Your instance
(_command/) is gitignored and never travels here; keep it in a private repository of your
own if you want its history.
- Claude Code with plugin support. The console fleet is a Claude Code plugin
(
tools/mods/mission-fleet). - Node.js for the Mission Control App and the docs viewer.
- Tailscale (optional) to reach the app from your phone.
- macOS or Linux is the primary platform: every gate has a bash form. On Windows the leak gates and the reference check also ship as PowerShell; a few newer tools are bash only.
- The superpowers plugin (recommended): the doctrine uses its brainstorming, test-driven
development and systematic debugging disciplines, and falls back to the written versions in
framework/without it.
| Folder | What lives there |
|---|---|
| framework/ | The doctrine: the constitution, the engineering standard, routing, continuity, design, remote access |
| .claude/skills/ | The 26 skills |
| tools/ | The gates and hooks, the mission-fleet plugin and the Mission Control App |
| LIFTOFF.md | The one-time guided setup |
Mission Control joins two lineages proven on a real multi-venture portfolio: a governance kit
(gate-locked phases, evidence before claims, deviation registers) and an orchestration method
(agent routing, context isolation, the learning loop). The seed lessons in
framework/learning-seed/ are that portfolio's real corpus, de-personalized, and the leak
gates keep it that way on every publish.
License: PolyForm Shield 1.0.0, see LICENSE. In plain words: you may use
Mission Control for any purpose, at work too, change it and share it, as long as whoever
gets a copy from you also gets the license, or its URL, and its Required Notice: line.
The one purpose it excludes is providing a product that competes with Mission Control, or
with any product its author provides using it. Selling or hosting it, or something built
from it, as a substitute for it competes, and so does giving such a product away. That
makes Mission Control source-available, not open source. Where these words and the
license differ, the license governs. For other terms, open an issue on this repository.


