What you type
14 tokensadd rate limiting to the public API and make sure existing clients keep working
- Your lines
- 2
- Added guidance
- 15
- Sections
- 8
Editor integration
Bring PromptFlowEngine prompt optimization directly into your development workflow. Describe a change, and the extension builds a structured prompt grounded in the relevant files, guards it for secrets, and helps you review the result.
AI coding assistants do better with a specification than with a sentence. The extension is built around one loop: Codebase, Prompt, AI, Codebase. You pick a scope, describe the change in your own words, and it produces a sectioned engineering prompt with the relevant code attached.
Files are filtered and scored for relevance to the task, so the prompt carries the handful of files that matter and not the repository. Before anything is sent to chat, a privacy guard scans for credentials and personal data and can block, redact, or ask.
When the assistant has edited your files, the review step opens Source Control and can build a review prompt from the modified files to check them against the original task. The extension is distributed as a .vsix package and is not on the Visual Studio Marketplace yet.
Practical example
The same restructuring the extension runs, generated by the engine when this page was built. In the editor, the relevant files from your workspace are added as a further section.
add rate limiting to the public API and make sure existing clients keep working
## Role Act as a senior software engineer. ## Objective Add rate limiting to the public API. ## Requirements - Make sure existing clients keep working. - Follow the conventions, naming, and structure already used in the codebase. - Handle invalid input and failure paths explicitly. ## Constraints - Do not change unrelated code, and do not add dependencies unless the task requires them. - Keep existing public interfaces and API contracts unchanged. ## Edge cases - Invalid, missing, and oversized input. - Unauthenticated, unauthorized, and rate-limited requests. ## Testing - Success responses: status code and response shape. - Validation and error responses. - Add tests for the new behavior, including failure paths. - Run the existing test suite and report the result. ## Acceptance criteria - The new behavior works as described in the objective. - All tests pass and no unrelated files are changed. ## Expected output - Return every change as a unified diff or as complete new and updated files. - List the design decisions and assumptions you made.
Computed by the engine when this page was built. Every added line is labeled as engine guidance in the interface, so you can delete what does not apply.
Project facts and relevant files are added to the prompt and labeled as coming from the codebase.
An "explain this" request gets no test plan. A bug fix gets edge cases and acceptance criteria.
Strict, redact, and warn modes for credentials and personal data.
A review prompt built from the changed files verifies the AI output against the task.
Four steps in the Activity Bar: Codebase, Prompt, AI, Codebase.
Entire workspace, current file, chosen files or folder, or no code.
Role, objective, context, code, requirements, constraints, edge cases, testing, acceptance criteria, and output.
Works on a selection or a document and opens a diff of original against restructured.
Verified, meaning-preserving removal of filler and repetition.
Registers the engine as MCP tools for Copilot agent mode on activation.
Select the workspace, describe the bug, and send a prompt that names the affected files and what must not break.
After the assistant edits files, build a review prompt from the uncommitted changes.
Run Restructure Prompt on a prompt stored in the repository and accept the diff.
Let Copilot agent mode call analyze, restructure, and sanitize through the bundled MCP server.
Not yet. It is distributed as a .vsix package that you build from the repository and install with the code --install-extension command.
No. The engine runs inside VS Code. Code leaves your machine only when you choose to send the generated prompt to an AI chat, after the privacy guard has checked it.
No. Send to AI Chat uses VS Code Chat when it is available and copies the prompt otherwise, so it works with any assistant.
No. The prompt is generated by deterministic rules, so the same task and files produce the same prompt.
Every product runs the same engine, so they combine without surprises.
Guides from the prompt engineering knowledge center that explain the ideas this product applies.
Free, deterministic, and private. No sign-up.