AVAILABLEAI Prompt Chain Library
Available product — files, demo URL and secure download get mapped in EDF 0.3.
What it does
Prompt pipelines that don't fall over.
Twelve tested prompt chains: pipelines of model calls and code where each step's output is checked before the next one runs. Blog post with a self-critique loop, long-document summary (map-reduce), meeting minutes, customer-feedback themes counted in code, email reply with tone check, interview questions, translation with a back-translation check, contract clause extraction with quotes verified in code, headline variants scored and picked, research brief with citations checked in code, lesson plan with a validated quiz, and support-ticket triage with rule-based routing. The engine validates every JSON reply against a schema, asks the model once to repair a bad reply with the exact problems listed, runs per-item map steps, critique loops that end early, code steps and quality gates that stop the chain instead of passing doubtful output on, enforces a hard cap on model calls, and fences untrusted text as data. Includes a quality gate for your own chains, adapters for OpenAI-compatible APIs and Anthropic, a CLI and a browser Studio where you read every prompt, edit the inputs and replay each eval case step by step. Node.js 22+, no npm dependencies, 58 automated tests. Eval replays use a scripted model, so they prove the chains and the engine are consistent, not how your model answers the prompts; the guide shows how to test with your own model.
Origin: EDF native product (Platform Modules 1.0) · Roadmap window:
Live demo


Package
Placed in the private vault layout (deploy/vault/products). Downloads stay disabled until checkout is connected.
What's included
- 12 chains across content, documents, operations, customer insight, communication, HR, localisation, marketing, research, education and support, each with inputs, a hard call cap, a final template and 4-6 eval cases (58 in total)
- Engine: JSON replies parsed (fenced or chatty is fine) and validated against a schema, one repair request listing the exact problems, then a clear failure; map, loop with early exit, code, gate and conditional steps
- Code steps for what models are bad at: counting, de-duplication, ranking, splitting long text, verifying quotes and citations against the source
- Safety: untrusted input fenced as data in every prompt that carries it (enforced by the quality gate), hard call limits the caller can only lower, bounded loops and lists
- Quality gate (lint) for your own chains: variables used before they exist, JSON steps without schemas, unfenced untrusted input, call cap below the worst case, too few evals, no safe-failure case
- Adapters for OpenAI-compatible APIs (OpenAI, Azure-style gateways, OpenRouter, vLLM, Ollama, ModelRouter) and Anthropic, tested against scripted HTTP responses
- Exports: Markdown with every prompt, JSON definition, OpenAI and Anthropic request bodies per step
- Studio in the browser: gallery, pipeline view, prompts rendered with your inputs, step-by-step replay of eval cases showing prompts, replies, rejected replies and repairs, loops and gates; CLI for list, show, prompts, export, validate, eval and run
- 58 automated tests, full source and docs: testing with your own model, writing chains
FORGE DNA™
TIMEPROOF SCORE — derived from real catalog fields (stack, formats, skill, AI flag), not decoration
Compatibility & relations
Derived by the mapping engine (family · layer · tags · dependency rules). Stored in the catalog as compatibility, dependencies, relations[].
Documentation
- Quick start (planned — every product gets one)
- Customisation guide
- Deployment guide — Vercel / Cloudflare / GitHub
- epic.json / EPIC-DNA.json — machine-readable DNA shipped inside the ZIP
Version history & changelog
| VERSION | STATE | NOTES |
|---|---|---|
| v1.0.0 2026-10-13 | AVAILABLE | Initial release. This package contains: 12 chains across content, documents, operations, customer insight, communication, HR, localisation, marketing, research, education and support, each with inputs, a hard call cap, a final template and 4-6 eval cases (58 in total); Engine: JSON replies parsed (fenced or chatty is fine) and validated against a schema, one repair request listing the exact problems, then a clear failure; map, loop with early exit, code, gate and conditional steps; Code steps for what models are bad at: counting, de-duplication, ranking, splitting long text, verifying quotes and citations against the source; Safety: untrusted input fenced as data in every prompt that carries it (enforced by the quality gate), hard call limits the caller can only lower, bounded loops and lists; Quality gate (lint) for your own chains: variables used before they exist, JSON steps without schemas, unfenced untrusted input, call cap below the worst case, too few evals, no safe-failure case; Adapters for OpenAI-compatible APIs (OpenAI, Azure-style gateways, OpenRouter, vLLM, Ollama, ModelRouter) and Anthropic, tested against scripted HTTP responses. |
| → | Roadmap | Target window for the next stage. |
License
Standard (personal or commercial use in one end project or your own business) · Extended ×3 (SaaS, unlimited client projects, finished products only). Agency licence comes later.
Deployment
Activated by 1.0 (download) and 2.0 (one-click deploy). Download will be a 10-minute signed URL from the R2 vault — never a public /files/product.zip.
Bundle & save
Related products
EPIC AgentOS

2 · configure
3 · deploy