Baseline Presets
For you if you run at least one agent in production and are about to build the next one — or to rewrite a first-generation prompt that grew by accretion. Also for the engineer who has just been told "we need a support agent by Friday" and already knows what a naive prompt does under real traffic.
one developer
For one developer. Deploy any preset as the system prompt of your own commercial product.
Get Solo — $39(opens in a new tab)up to 10 seats
One flat price for up to 10 engineers in one organisation — expense it without a PO or per-seat billing.
Get Team — $149(opens in a new tab)Prices are in USD. Applicable VAT or sales tax is calculated and added by Gumroad at checkout.
Every team running agents ends up writing the same system prompts: a support agent that can touch orders, a docs assistant that has to cite, a SQL agent that must never write, a router in front of all of them. The first version is always plausible and always missing the same things — the carve-outs that stop refusals from blocking legitimate requests, the confirmation step before a tool changes something, the output contract downstream code depends on. Baseline Presets is 15 of those prompts, written properly once: each with template slots, reference tool schemas, a JSON output contract and 12 contract tests, so you start from something that already holds and find out the day an edit breaks it. Prompts, schemas and tests only — you wire them into your own stack.
- Guarantee
- 14-day money-back guarantee. Email maksim344551@gmail.com within 14 days of purchase and we'll refund your full order through Gumroad, no questions asked.
- Delivery
- Instant download from your Gumroad library right after purchase. Updates to a pack you own are free and arrive the same way: re-download it there for the latest version.
- Release
- v2026.09 · updated 2026-09-29
01What's inside
- Customer-Facing Agents (3) — support agent with gated account actions, docs Q&A with enforced citations, read-mostly in-app copilot
- Data & Operations Agents (3) — read-only SQL analyst, web research agent with a source policy, ticketing & workflow operator
- Engineering Agents (3) — PR code review bot, GitHub issue triage bot, CI failure explainer
- Routing, Orchestration & Memory (3) — front-door intent router, orchestrator with worker brief, context compaction summarizer
- Structured Extraction (3) — schema-driven document extractor, inbound email & ticket parser, transcript to decisions & action items
- Shared policy blocks — untrusted-content envelope, irreversible-action confirmation, ambiguity lanes, AI disclosure & human handoff, output repair rule; each preset inlines the ones it needs, so it works when pasted alone
02Why it holds up
- Every preset uses the same section order and instruction hierarchy, so your agents behave consistently and a fix in one carries over to the rest.
- Each ships with reference tool schemas, a JSON output contract with a repair rule, and 12 contract tests in JSONL — load them into promptfoo, pytest, Vitest or your own runner.
- Every preset states what has to be enforced in code rather than trusted to the prompt.
- Drafted by an AI model, then checked by scripted structure checks and one adversarial AI review pass. No human expert reviewed it, and no preset was run against a model. The contract tests are synthetic and are there for you to run on your own model and data.
How it's made The packs are drafted and reviewed with AI models and curated by the author. More on the About page.
03What this is not
- Not agent code, a framework or an eval runner — prompts, schemas and test files you wire into your own stack.
- Not benchmarked: the contract tests are for you to run on your model and data, and we don't publish scores.
- Not a compliance product: the AI-disclosure slot supports EU AI Act Art. 50(1)-style disclosure, but it doesn't make your system compliant, and regulated domains (medical, legal, financial advice) need your own domain review.
Going further
Once a preset is live, Redline Prompts is how you audit it against your real traces, and Press Check is how you grade its output at scale. An Agent Harness + Evals Kit that wires both into CI is in development; it is not available yet.
04Questions about Baseline Presets
01Can't I just ask a model to write a system prompt?
You can, and you'll get something plausible. What you usually won't get on the first try: carve-outs so refusal rules don't block legitimate adjacent requests, a confirmation protocol for tools with side effects, a machine-checkable output contract with a repair rule, the list of controls that belong in code rather than the prompt, and a test file that catches the day an edit breaks one of them. That's what each preset is.
02Which models and frameworks does it work with?
The presets are model-agnostic. Tool schemas are plain JSON Schema, with notes for Anthropic tool use, OpenAI function calling and Gemini function declarations, and every preset names the model tier it is written for and what to restate on small, fast models. We don't publish benchmark scores; the contract tests are there so you can check a preset on your own model.
03Can I ship a preset as the system prompt of my commercial product?
Yes — that's what it's for. Deploying a preset inside your own product is covered by the license, and a user extracting the prompt from your deployed agent isn't a breach on your side. What the license rules out is reselling or republishing the pack itself.
04How is this different from Redline Prompts?
Redline audits and diagnoses prompts you already have; Baseline gives you the prompt to start from. Each preset names the Redline prompts worth running once you've customized it, and the Press Check judges that grade its output — but you don't need either to use Baseline.
05What exactly do I download?
A .zip with baseline-presets.md (all 15 presets, readable end to end) and one folder per preset containing system-prompt.md, tools.json, output.schema.json and contract-tests.jsonl, so you can diff, version and test them like code. Updates are free and arrive the same way: re-download from your Gumroad library.
05Reviews
Visitor rating
No rating yet
Unavailable
- 5 stars:0 reviews
- 4 stars:0 reviews
- 3 stars:0 reviews
- 2 stars:0 reviews
- 1 star:0 reviews
Posted by site visitors and checked by hand before they appear. Purchases aren't verified.
Leave a review
Reviews are temporarily closed.
06More packs
Redline Prompts
50 expert-grade prompts for the agent engineering work you can't wing — system prompts, evals, failure diagnosis, injection hardening, and 6 more categories.
Press Check
15 ready-to-run LLM judges for agent output — groundedness, tool-call provenance, side-effect confirmation, injection compliance, refusal calibration and 10 more — each with anchored labels, strict JSON output and a gold set.