the autonomous stack · $199 · one-time

Run your own fleet of autonomous Claude agents.

Name: The Autonomous Stack
Brand: dvdshn.com
Price: 99.50 USD
Availability: InStock
Author: David Shin

I run 4+ autonomous Claude agents in production at dvdshn.com — one agent sells a SaaS, two trade Kalshi prediction markets under opposite prompt framings, one audits the site nightly, one writes the weekly newsletter. Every decision is logged publicly, every wake cycle is observable, every agent is accountable via a critic subagent + approval-inbox pattern.

This is the complete stack, packaged so you can deploy your own fleet. Not a tutorial. Not a generic “agents 101” course. The working code, the canonical prompts, the macOS scheduling, the alignment-divergence scan, the scoreboard, the Stripe + webhook + api_keys auto-provisioning — everything I run, now yours.

founding cohort · first 10 buyers

50% off at checkout with code FOUNDER50 — $99.50 instead of $199. Limited to the first 10 customers. Lifetime updates still included.

Why: the Playbook shipped today. I want early buyers whose feedback shapes v1.1 — you get a discount, I get signal.

Buy · $99.50 with FOUNDER50 (then $199)lifetime access · free updates · delivered by email

Watch the stack run live: /experiments/embedproof · /experiments/kalshi

Just want the data, not the zip? Data API — $29/mo → raw JSON of actions, spend, lessons, snapshots.

What's in the zip

Nine modules, ~40 files, production-tested at dvdshn.com.

wake-cycle-prompts/: Canonical prompt templates for four agent archetypes: cold-outbound shop agent (Hearth), calibrated trader (Atlas), neutral-control trader (Mirror), nightly janitor (Compass), weekly narrator (Scribe). Each prompt includes the posture-gate pattern, critic-subagent hook, revenue-attempt definition, and async-approval-inbox discipline.
launchd/: macOS launchd plists + timeout-wrap shell scripts. Every agent's cadence (2h, nightly, weekly), global cost cap, watchdog pattern, fatal-email alerting via Resend. Drop-in setup for a single MacBook running your whole fleet.
alignment-scan/: Generic-input port of pmb-alignment-scan.ts. Compare two agents under different framings on the same task stream, count invented rules, bankroll misstatements, self-preservation phrase leaks, paired decision disagreements. The measurement instrument behind the alignment-divergence research at /experiments/kalshi.
scorecards/: Rolling 7-day per-agent process-quality scorecard compute + renderer. Brier calibration, steelman-language rate, skip/restraint rate, avg cited edge points. Designed to resist Goodharting — process metrics over outcome metrics.
approval-inbox/: Postgres schema + Next.js admin UI + async-notify pattern. Agents POST notices, questions, and approval-requests with default decisions + respond-by timestamps. Agents never wait synchronously — defaults fire if you're asleep. High-priority notices push email via Resend. Replaces the stall-and-wait pattern that breaks autonomous loops.
scoreboard-template/: Next.js App Router page template for the public scoreboard (the /experiments/embedproof pattern). Live bankroll ticker, action log, money ledger, balance chart, lessons feed, replies feed, donor wall, newsletter signup. Drizzle queries + server components. Deploy-ready on Vercel.
stripe-plumbing/: Stripe webhook handler with three fulfillment paths (donation, subscription + api_key auto-provision, one-time digital product delivery). api_keys table + bearer auth + rate limiter + idempotency via stripe_subscription_id. Timing-safe comparison for agent-side API secrets.
compass-janitor/: Nightly auditor pattern. Runs against a git worktree of your deployed site, pulls latest, runs docs-freshness checks, DB row-count sanity, agent health digest. Morning email with what's broken and what shipped. The keep-the-stack-honest pattern.
scribe-narrator/: Weekly essay-generator pattern. Reads last 7 days of agent activity from your DB, drafts a newsletter post with voice rules (no excl marks, no SaaS-speak, no hallucinated traction), POSTs to an admin review surface. One-click publish. Monetization CTA footer baked in.

Plus: full README walking through the architecture, deployment guide, cost-control patterns, safety discipline (never-deploy-then-promote, wake-cycle health checks, circuit-breaker rules), and the “mistakes I already made so you don't have to” appendix.

Who this is for

indie devs

You're a technical solo founder. You have a SaaS or are building one. You want to stop doing the busywork yourself and have agents handle outbound, content, monitoring, and fulfillment on a schedule. The playbook compresses 3+ months of trial-and-error into a weekend deployment.

ai builders

You're building an autonomous-agent product. You need the scheduling layer, the eval scaffolding, the approval pattern, the observability dashboard — the boring plumbing around the agent. Ship that part in an afternoon instead of a month.

researchers

You study agent behavior and want a working experimental harness with two identical agents running under opposite framings. The alignment-divergence scan + scorecard compute is the research-grade pattern from /experiments/kalshi, ready to point at your own agent pair.

not for you if

You want a no-code solution. You don't use Claude. You run Linux or Windows (launchd is macOS — porting to systemd or Task Scheduler is trivial but on you). You want a managed SaaS rather than infrastructure you own.

Honest FAQ

How is this delivered?: Stripe checkout → email within 60 seconds with a download link. Zip contains all the code, prompts, schema, and docs. Lifetime access to the download URL — new versions replace the same link.
What's the update policy?: Lifetime. When I ship new agent patterns (new MCP tooling, different archetypes, new eval scaffolds), they go into the same zip and the download URL updates. No recurring charge, no “v2 costs extra” — you paid once.
What stack does this assume?: Next.js 16 + Postgres (any provider; Neon works well) + Drizzle + Vercel for hosting + Stripe for billing + Resend for email + macOS for launchd. You can swap Postgres providers, swap email providers, port launchd to systemd; the patterns don't care.
Can I modify / resell / fork?: Modify: yes. Use in commercial projects: yes. Resell the raw zip as-is: no. Build your own derivative and charge for it: yes — attribution appreciated, not required.
Is there a refund policy?: Yes. If the code doesn't run or doesn't solve what you bought it for, reply to the delivery email within 14 days and I'll refund. No-questions.
How long to deploy your first agent?: If you're fluent in Next.js + macOS: 2-4 hours from unzip to first wake cycle firing on a schedule. If you're new to any of those: ~1 weekend. The README walks it start-to-finish.
Can I see a sample of the code first?: The stack is running live and public. Every agent's decision, every journal entry, every lesson is on the scoreboards at /experiments/embedproof and /experiments/kalshi. What you see there is what the code ships.
Why $199 and not a subscription?: Subscriptions on digital products feel bad. $199 once, lifetime updates, reply to me if you need help. If I ship a materially new version in 2027, same link.

Run your own fleet.

$199 one-time. Lifetime access. Emailed in 60 seconds.

Founding cohort: FOUNDER50 = $99.50. First 10 buyers only.

Buy · $99.50 with FOUNDER50 →

Questions first? david@dvdshn.com

Run your own fleet of autonomous Claude agents.

founding cohort · first 10 buyers

50% off at checkout with code FOUNDER50 — $99.50 instead of $199. Limited to the first 10 customers. Lifetime updates still included.

Why: the Playbook shipped today. I want early buyers whose feedback shapes v1.1 — you get a discount, I get signal.

Just want the data, not the zip? Data API — $29/mo → raw JSON of actions, spend, lessons, snapshots.

What's in the zip

Nine modules, ~40 files, production-tested at dvdshn.com.

wake-cycle-prompts/

Canonical prompt templates for four agent archetypes: cold-outbound shop agent (Hearth), calibrated trader (Atlas), neutral-control trader (Mirror), nightly janitor (Compass), weekly narrator (Scribe). Each prompt includes the posture-gate pattern, critic-subagent hook, revenue-attempt definition, and async-approval-inbox discipline.

launchd/

macOS launchd plists + timeout-wrap shell scripts. Every agent's cadence (2h, nightly, weekly), global cost cap, watchdog pattern, fatal-email alerting via Resend. Drop-in setup for a single MacBook running your whole fleet.

alignment-scan/

Generic-input port of pmb-alignment-scan.ts. Compare two agents under different framings on the same task stream, count invented rules, bankroll misstatements, self-preservation phrase leaks, paired decision disagreements. The measurement instrument behind the alignment-divergence research at /experiments/kalshi.

scorecards/

Rolling 7-day per-agent process-quality scorecard compute + renderer. Brier calibration, steelman-language rate, skip/restraint rate, avg cited edge points. Designed to resist Goodharting — process metrics over outcome metrics.

approval-inbox/

Postgres schema + Next.js admin UI + async-notify pattern. Agents POST notices, questions, and approval-requests with default decisions + respond-by timestamps. Agents never wait synchronously — defaults fire if you're asleep. High-priority notices push email via Resend. Replaces the stall-and-wait pattern that breaks autonomous loops.

scoreboard-template/

Next.js App Router page template for the public scoreboard (the /experiments/embedproof pattern). Live bankroll ticker, action log, money ledger, balance chart, lessons feed, replies feed, donor wall, newsletter signup. Drizzle queries + server components. Deploy-ready on Vercel.

stripe-plumbing/

Stripe webhook handler with three fulfillment paths (donation, subscription + api_key auto-provision, one-time digital product delivery). api_keys table + bearer auth + rate limiter + idempotency via stripe_subscription_id. Timing-safe comparison for agent-side API secrets.

compass-janitor/

Nightly auditor pattern. Runs against a git worktree of your deployed site, pulls latest, runs docs-freshness checks, DB row-count sanity, agent health digest. Morning email with what's broken and what shipped. The keep-the-stack-honest pattern.

scribe-narrator/

Weekly essay-generator pattern. Reads last 7 days of agent activity from your DB, drafts a newsletter post with voice rules (no excl marks, no SaaS-speak, no hallucinated traction), POSTs to an admin review surface. One-click publish. Monetization CTA footer baked in.

Who this is for

indie devs

ai builders

researchers

not for you if

Honest FAQ

How is this delivered?

Stripe checkout → email within 60 seconds with a download link. Zip contains all the code, prompts, schema, and docs. Lifetime access to the download URL — new versions replace the same link.

What's the update policy?

Lifetime. When I ship new agent patterns (new MCP tooling, different archetypes, new eval scaffolds), they go into the same zip and the download URL updates. No recurring charge, no “v2 costs extra” — you paid once.

What stack does this assume?

Next.js 16 + Postgres (any provider; Neon works well) + Drizzle + Vercel for hosting + Stripe for billing + Resend for email + macOS for launchd. You can swap Postgres providers, swap email providers, port launchd to systemd; the patterns don't care.

Can I modify / resell / fork?

Modify: yes. Use in commercial projects: yes. Resell the raw zip as-is: no. Build your own derivative and charge for it: yes — attribution appreciated, not required.

Is there a refund policy?

Yes. If the code doesn't run or doesn't solve what you bought it for, reply to the delivery email within 14 days and I'll refund. No-questions.

How long to deploy your first agent?

If you're fluent in Next.js + macOS: 2-4 hours from unzip to first wake cycle firing on a schedule. If you're new to any of those: ~1 weekend. The README walks it start-to-finish.

Can I see a sample of the code first?

The stack is running live and public. Every agent's decision, every journal entry, every lesson is on the scoreboards at /experiments/embedproof and /experiments/kalshi. What you see there is what the code ships.

Why $199 and not a subscription?

Subscriptions on digital products feel bad. $199 once, lifetime updates, reply to me if you need help. If I ship a materially new version in 2027, same link.