Forgehaven Labs · The Autopsy
One solo founder, an AI agent fleet, and a portfolio of real products. Every month we publish the engineering teardown: the real numbers, what got killed, what survived, and why. No hype, no income claims; wins and failures both make the cut. The kit that runs the audit is $29.
Monthly cadence · pre-committed kill gate · every number verified against code and git
Post #1 · publishing soon
The ledger first, because everything after it is commentary. These numbers come from a 45-agent self-audit run on 2026-07-01: 17 ground-truth repo audits, 8 market-reality research reports, 14 skeptical judges forced to output probabilities, and 1 red-team whose only job was to attack the judges' numbers.
| Metric | Value |
|---|---|
| Product lines built or substantially advanced | 17 |
| Wall-clock time | ~2 months of nights and weekends |
| Repos | 43 |
| Passing tests across the portfolio | ~1,900 |
| Real bugs the agents caught reviewing their own code | a stored XSS, a payment-link injection, an export over-count, a banned-word homograph in a kids' story generator |
| Products that survive the audit | 2, plus a services funnel |
The full post opens the complete revenue ledger, every figure and nothing off the books, then covers the zombie jobs still running for a product that had already been killed, the probability table for all 17 products, and the four takeaways. It publishes here and on the usual channels once the payment links and click measurement are live; that is its own rule from the post.
The series
Because the audit found that the portfolio's real deficiency was never build capacity; it was distribution and the nerve to publish real numbers with a name attached. So the autopsy is a monthly habit: numbers first, verdicts in writing, gates that actually get executed.
Every post opens with the ledger. Revenue, customers, tests, kills. If a number cannot be verified against code, git history, or a live URL, it does not get printed.
Pre-committed, in writing: two consecutive monthly posts with zero funnel clicks and the series drops to quarterly, permanently, no relitigating. Gates apply to content too.
When a product dies, the jobs get unloaded, the keys revoked, and the repo archived the same week. Zombies charge rent.
An agent fleet moves your bottleneck, it does not remove it. Count bets by correlation, not repo count. A product without a checkout URL is a hobby with tests. Execute kills the same week.
The product
The actual prompts and templates from the 45-agent audit, sanitized and packaged. Not a course. Most solo-founder portfolio reviews fail because the founder grades their own homework; this kit splits the review into adversarial layers so the output is a scoreboard you can act on.
$29
One-time. The working files behind the series.
Receipts from the audit that built this page:
Requirements: Claude Code (or any agent CLI), your repos on local disk, 2 to 6 hours of mostly unattended wall clock, and the stomach to read the output. Personal license; run it on your own and your clients' projects, do not resell the kit itself.
Prefer it done for you?
The Overnight Audit is the same machinery pointed at one repository: an agent fleet runs the passes, a human verifies every finding, and you get a severity-ranked report with a 30-day fix plan inside 24 hours.