Claude Code sessions forget everything. Your agent shouldn't.
A battle-tested folder scaffolding that turns Claude Code into a self-running, self-remembering autonomous operator — extracted from a real system that runs itself every day.
Autonomous agents don't fail loudly. They rot quietly.
Most "autonomous agent" setups fall apart within days — and every failure is one of these four:
amnesia
Every session starts from zero. Context re-explained, work re-done, boot cost grows forever.
drift
Without standing orders, the agent slowly wanders off-mission and nobody notices for weeks.
silent stall
An unattended run hits a permission prompt and just… stops. You find out days later.
overreach
The agent does something you never sanctioned — because nothing defined what needs your sign-off.
The missing infrastructure layer
Agent Autopilot is not a prompt pack. It's the operating scaffolding a scheduled Claude Code agent needs to survive weeks of nobody watching:
Persistent memory
A snapshot file rewritten every session plus append-only history — boot cost stays constant whether your agent is 3 days or 3 months old.
A constitution
Standing orders with hard invariants the agent can never weaken, and a self-modification protocol for everything else — so it gets better over weeks, not just older.
Human-in-the-loop interface
A two-file mailbox plus desktop popups, with a strict money/credentials protocol: the agent never spends a cent or touches an account without routing exact steps through you.
Ops that survive nobody watching
Scheduling via the desktop app (gotchas documented), permission pre-approval workflow, model/usage fallback policy, git as a time machine — plus desktop notifiers for Windows, macOS and Linux that can never hang a session waiting on a click (new in v1.2).
Visibility without babysitting
A one-page status dashboard (plain HTML, no server) the agent refreshes every session: mission KPIs, the latest session report, and an amber box for anything waiting on you.
Extracted from a real system — by the AI that runs it
This scaffolding operates a real autonomous agent every single day: same files, same boot and shutdown rituals, same failure-mode countermeasures. The kit is that system, generalized — and the generalization was written by the very AI that lives inside it.
No theory, no hype, and no income promises. This is infrastructure, not a get-rich course.
What's inside
Works with any mission you can write a paragraph about — an autonomous side-business, research analyst, codebase gardener, content pipeline, ops bot. Requires the Claude Code desktop app (any paid plan).
- Complete 16-file scaffolding kit
- The Manual: setup → unattended ops
- 10 failure modes, each with its fix
- Money & credentials safety protocol
- Status dashboard — new in v1.1
- Cross-platform notifiers — new in v1.2
- Future v1.x updates included
Questions
What exactly do I need to run this?
The Claude Code desktop app on any paid plan. The kit's scheduling instructions use the desktop app's built-in scheduled tasks. Windows, macOS, and Linux are all genuinely covered — as of v1.2 the kit ships real notifier scripts for both Windows and macOS/Linux, not documented one-liners.
Is this a "make money with AI" course?
No. It contains no income claims and no business advice. It's the infrastructure layer — memory, constitution, human interface, and operations — for whatever mission you point your agent at.
How is this different from just writing a good CLAUDE.md?
A CLAUDE.md tells one session what to do. This kit makes hundreds of sessions behave like one continuous operator: state snapshots rewritten every session, append-only decision and journal logs, boot/shutdown rituals, hard invariants, and a protocol for everything that needs a human. That's the part that keeps an agent alive in week six.
Do I get updates?
Yes — v1.x updates are included. Gumroad notifies you when the file changes.
Is this affiliated with Anthropic?
No. Not affiliated with or endorsed by Anthropic. "Claude Code" is referenced only to describe compatibility.