Documentation

PromptFlowEngine docs

PromptFlowEngine is one deterministic TypeScript engine (@promptflow/engine) that parses, analyzes, restructures, optimizes, verifies, and scores prompts. The web workspace, Chrome extension, VS Code extension, MCP server, HTTP API, and CLI are all interfaces to that same engine, so a prompt gets the same result everywhere. New here? Start with getting started.

Learn

• Prompt engineering guides26 guides on writing, optimizing, testing, and securing prompts.
• Prompt library40 reusable prompts in 20 categories, from code review to research.
• Prompt templates6 production prompt templates with {{variables}}, each scored strong.

Core concepts

• How the engine worksThe pipeline, the rule catalogue, the scoring formula, and candidate verification.
• Models & pricingWhich models the token counts and cost estimates cover, and how prices are kept current.

Integrations

• Connect PromptFlowEngineCompare the extensions, MCP, and the HTTP API, with setup, verification, and troubleshooting.
• VS Code extensionCodebase → Prompt → Codebase in the editor: generate, restructure, send, and review.
• Chrome extensionAnalyze, restructure, and redact prompts inside ChatGPT, Claude, Gemini, Grok, Perplexity, and DeepSeek.
• MCP clientConnect Copilot, Claude, Cursor, or any MCP client: tools, resources, and configuration.
• HTTP APIStateless JSON endpoints, limits, errors, and examples.
• CLI & CI gateEvery command, plus the check gate for prompt files in CI.

Security

• Privacy & securityWhat runs where, what is stored, and what is logged.

Guides by topic

The prompt engineering guides explain the techniques the engine's rules check for. Each topic below opens its section of the guide index.

• Fundamentals (5): What prompt engineering is, how to do it step by step, and the habits that separate a prompt that works once from one that works every time.
• Prompt types (4): The basic shapes a prompt can take: with or without examples, with a role, as a system prompt, and around a required output format.
• Techniques (5): Methods for harder tasks: reasoning step by step, splitting work across several prompts, giving a model tools, and improving a prompt by measurement.
• Optimization, testing, and security (6): Running prompts in production: token and context budgets, evaluation, versioning, injection defenses, and keeping secrets out of prompts.
• AI agents and MCP (3): Prompts that drive tools instead of a single reply: agent instructions, retrieval, and the Model Context Protocol.
• Model guides (3): What changes between providers: how OpenAI, Anthropic, and Google models differ in the way they read instructions, context, and formats.

Prompt library and templates

The prompt library holds single-purpose prompts you can copy and adapt. The developer categories are a good place to start:

• Software Development prompts: Prompts for planning changes, reviewing code, writing tests, and debugging, structured so an assistant returns work you can check.
• AI Engineering prompts: Prompts for building with models: designing evaluations, writing tool definitions, and planning retrieval pipelines.
• Prompt Engineering prompts: Prompts about prompts: critiquing a draft, generating test cases, and converting a request into a structured system prompt.
• DevOps prompts: Prompts for delivery and operations: reviewing pipelines, writing runbooks, and hardening container images.
• Research prompts: Prompts for reading and comparing sources: paper summaries that separate findings from claims, and literature comparison tables.
• Productivity prompts: Prompts for everyday work: turning notes into a plan, prioritizing a task list, and drafting clear email replies.

Prompt templates are longer, reusable starting points with {{variables}} for the parts that change: pull request review, customer support reply, structured data extraction, meeting summary, ticket classification, and tool-using agent system prompt.

The core workflow

Codebase → Prompt → AI → Codebase. Start from the code you have, turn a rough request into a structured engineering prompt grounded in that code, send it to the model you already use, then review what comes back against the task. The VS Code extension runs the whole loop; the other products cover the prompt step wherever you happen to be working.

Products

These pages are the reference. For an overview of each product, see the AI prompt tools: the prompt analyzer, prompt optimizer, prompt security scanner, prompt testing, and prompt evaluation.

What it does not do

• It does not call a language model. Every result comes from documented rules, so it is free, instant, private, and reproducible.
• It does not invent missing content. Gaps such as an unspecified output format become suggestions for you.
• It does not run your prompt against a model. Model-based testing is on the roadmap.