What’s live, what’s open source, and what stays private

Start with the field work: systems deployed into billing, operations, and media workflows. Then inspect the public extracts and demos. Deeper backend checks live in the demo operations page.

Field proof

The production repositories stay private. Each case study names the operator problem, deployment state, evidence basis, and public extract where one exists. Case studies

Meter

Private production system

Status: Running in an MSP billing stack

Evidence: In production for an MSP; one reconciliation audit covered 202 accounts and 931 work items (self-reported)

Throughline

Private production integration

Status: Shipped and actively maintained

Evidence: Six vendor integrations share one connector contract; architecture notes estimate 69% fewer daily API calls

CoHost AI Studio

Private applied-AI system

Status: Live, pre-launch

Evidence: A 23-step media pipeline with 11 programmatic scorers, five optional LLM evals, and a human release gate

Open source

Every entry names the file worth reading first. Entries tagged cold-clone reviewed were verified by an independent engineer working from a fresh clone before the repository went public. 27 public extracts

Shipwright

Cold-clone reviewed

What it is: An agent that takes an approved GitHub issue and returns a tested, reviewable pull request.

Read first: The agent runner's test suite, which exercises the plan, patch, and review loop end to end.

Open the runner tests · View code

Holdfast

Cold-clone reviewed · 38 tests

What it is: An append-only decision ledger with a human publish gate, enforced inside Postgres by triggers, a hash chain, and unique indexes.

Read first: Attack tests that try to rewrite history and race two sessions against the publish gate; the migrations under them close what the tests document.

Open the attack tests · View code

Vouch

Cold-clone reviewed · 173 tests

What it is: Human review as an API: agents submit work, real reviewers return a consensus verdict.

Read first: A README walkthrough of the offline verification harnesses, with a 173-test suite behind it.

Verify it yourself, offline · View code

Matchbox

Cold-clone reviewed · 342 tests

What it is: Disposable cloud servers with a lease. When it expires, teardown writes a receipt proving the machine is gone.

Read first: A hermetic reaper suite covering lease expiry, escalation, and the receipts a destroyed server leaves behind.

Open the reaper tests · View code

Tenantwell

Cold-clone reviewed · 30 tests

What it is: Postgres row-level-security multi-tenancy, tested by an adversarial suite that attempts cross-tenant reads.

Read first: SQL attack tests that probe the tenant boundary from inside a session, including the one documented way across it.

Open the isolation tests · View code

Reconciler

What it is: Finds billing discrepancies, proposes typed fixes, and requires a person to approve every invoice change.

Read first: The verification workflow CI runs on every push, plus a README that walks the local run.

Open the verification workflow · View code

Landfall

47 tests

What it is: Migrates a WordPress export to Astro and fails the build when any origin URL is missing from the built site.

Read first: Parity tests plus the synthetic Harborline fixture the verifier runs against. No client URLs are in the repository.

Open the parity tests · View code

RepoMan

Grades a codebase's technical debt and prices the paydown.

Open CI · Repo

Marrow

Turns AI coding-session transcripts into durable, searchable lessons.

Open the isolation test · Repo

Winnow

Scores and routes inbound leads; a 24-case model eval suite gates every accuracy change.

Open the eval tests · Repo

Muster

A terminal dashboard for running a fleet of AI coding agents in tmux.

Open CI · Repo

Broadside

Fans one prompt out to a fleet of AI coding agents in tmux panes, then collects their work.

Open the cleanup-safety test · Repo

Yardmaster

One CLI that routes a prompt to Claude, Codex, Cursor, or Gemini, with JSON output and real exit codes.

Open the exit-code tests · Repo

Pairwell

PostgreSQL fuzzy matching for public filings, with hermetic SQL assertions and a bounded live demo on Texas open data.

Open the SQL assertions · Repo

Deckhand

An autonomous coding agent that works a Linear backlog end to end and opens the pull requests.

Open the security tests · Repo

Throughline Connector Kit

The sanitized public kit behind Throughline: one four-method connector contract, the sync engine, sample CRM data, and tests.

Open the contract tests · Repo

Trifold

One Zod-validated operation registers as an HTTP route, a CLI command, and an opt-in MCP tool, with OpenAPI emitted from the same registry.

Open the registry tests · Repo

Vaultkeep

Reads, injects, and exports 1Password secrets from the command line so they never have to sit in a committed .env.

Read the 30-second start · Repo

Gradewell

Scores Claude prompts against a stored eval suite, with string-match or rubric grading you can re-run.

Open the compare tests · Repo

Onward

A Claude Code stop hook that sends the agent back to work when the last turn still lists next steps, and stops when it is asking you a question.

Open the stop hook · Repo

Placard

Forty-nine MCP tools over Meta's Marketing API, with a smoke test that starts the server and lists every tool without credentials.

Open the smoke test · Repo

Showwright

Compiles a JSON show definition into a vMix preset bundle a studio machine can open.

Open the demo episode · Repo

Dowser

Suggests the next Claude Code skill from a local library using fuzzy search, TF-IDF, and optional embeddings.

Open the analyzer hook · Repo

polycast

One command definition writes Raycast, PopClip, Dropzone, and Shortcuts artifacts, and will not emit a surface that cannot represent that command's input.

Open the emitter registry · Repo

xair-cli

Drives Behringer X Air and Midas MR mixers from the terminal, and a parity test fails the build if the docs drift from the CLI.

Open the doc parity test · Repo

Outfitter

Claude Code plugin packs for building, testing, and deploying, where CI runs the plugin validator over every pack on every push.

Open the pack validation · Repo

llm-1password

Resolves llm model keys from 1Password at prompt time. refs.json stores the op:// reference; a failed read aborts the call.

If a row here is the reason you want to talk, email me at [email protected].