Small, verified systems you drop into an AI pipeline.
Not a prompt marketplace. Each entry is a working module with a stated outcome, a test suite you can run in one command, an honest list of what it cannot do, and a licence you can read in a minute.
What these are
3 of these systems came out of one publishing pipeline that was already running them in production. The other 3 are existing products listed here because they belong in the same catalogue, not because they were rebuilt for it.
| Question | Answer |
|---|---|
| What is it? | A directory of files you copy into your own project. No SaaS, no account, no runtime service. |
| Who is it for? | Solo operators and small teams running automated or semi-automated content and acquisition pipelines, and engineers adding guardrails to an LLM step. |
| What does it need? | Node.js 18 or newer. No package installs, no build step, no network access, no telemetry. |
| How hard is it? | Low. The work is not installation, it is deciding what your contract, catalogue or channel list actually says. |
| Is it documented? | Every package ships a README with setup, configuration, sample input and output, known limitations, failure handling, rollback and security notes. |
| How do updates work? | Semver, a changelog per package, published as new versions. Nothing auto-updates and nothing phones home. |
Packaged systems
Free, MIT, and downloadable from this site. Each one is a directory of files with a README, examples and a test suite you run in one command.
-
Truth Gate
Stops an AI publishing pipeline from inventing a life it did not live: unevidenced first-person claims, fabricated biography and private data are refused with a named safe rewrite.
Free · MIT guardrails setup: low 14 tests -
Duplicate Guard
Stops your own content from competing with itself: source echo, copy-spin, literal translations, repeated headlines and two live items chasing one audience with one CTA.
Free · MIT guardrails setup: low 16 tests -
Attribution Contract
One routing table, identical tracked links every time, a CI check that fails when a CTA would arrive unattributed, and a funnel that reports NOT_MEASURED instead of inventing a rate.
Free · MIT measurement setup: low 18 tests
Paid
3 entries, none of them new. These are existing Stratum Praxis products listed here because they belong in the same catalogue.
-
AI Workflow Operator Bundle
Three operating systems for workflow architecture, AI/tool routing and company knowledge — with ownership, failure handling, QA and human approval boundaries written in.
$17 USD operating-systems setup: none -
Cross-Agent Operating Kit
Keep Brain, Policy, Skills and State portable across Claude Code, Codex, Cursor and future agent runtimes, with human gates, cost guards and a migration path.
$69 USD operating-systems setup: low -
AI Operations Systems Pack
Ten operating systems for running AI work as an operation rather than a series of one-off requests.
$39 USD operating-systems setup: none
Their pages, prices, checkouts and buyer delivery already existed and were not created, changed or duplicated for this library. Buying goes through the same checkout each has always used.
Free, personal, commercial and agency use
The short version, because this is where most catalogues are vague on purpose.
The three packaged systems are MIT. Personal use, commercial use inside your business, and client work are all permitted, at no cost, with no separate tier to buy. Keep the copyright notice; do not resell the source as your own product.
That is not a launch promotion. These modules were extracted from a public, MIT-licensed repository, so those files are already MIT to everyone. Selling a restrictive licence over them would not be honest or enforceable. The licence page explains it in full.
The paid entries keep their own terms. Each is sold under the terms published on its own page, with its own delivery behind payment verification. Those prices and terms were not created or changed here.
Writing
Published to this section rather than to a product page. Kept here so it stays reachable from the library it was written for.
- How to Run an AI Swarm Without Losing Control
A systems architect's guide to coordinating multiple AI agents without losing the ability to say what any one of them did.
Considered and not shipped
Five internal assets were audited and left out. Catalogue size is not the goal.
| Asset | Why it is not here |
|---|---|
publisher-gate | Four-file policy coupling; clean-room cost is high and the product is hard to explain without the surrounding engine. |
signal-intelligence | Inseparable from a populated asset inventory; packaging it would ship an empty scoring engine. |
revenue-publisher | Needs an owner decision: it calls a third-party AI API, which puts the buyer's cost and terms-of-service position on the seller. |
github-actions-starter | Nothing survives genericisation; it would be new writing presented as an extraction. |
prompt-pack | Duplicate of the already-live AI Workflow Operator Bundle. |