AI Dev Tasks

AI Dev Tasks (ai-dev-tasks)

by Ryan Carson / snarktank

Structured prompts and a simple 3-file system to guide AI-assisted feature development from PRD → tasks → tests.

See https://github.com/snarktank/ai-dev-tasks

Overview

AI Dev Tasks is a lightweight, practical repository of markdown prompt templates and a workflow designed to “supercharge” feature development when using AI-powered IDEs and CLIs (Cursor, Amp, Claude Code, Windsurf, Copilot-style assistants, etc.). The core idea is to replace ad-hoc, monolithic prompting with a repeatable, verifiable process: generate a Product Requirement Document (PRD), break it into atomic implementation tasks, then run and verify each task (tests and manual review) as the AI implements them.

The project is creator-focused — it targets solo founders and small teams who want to reliably build production features with AI assistance while maintaining control and quality.

Core components (the 3-file system)

The repository centers on three conversational/template files you drop into your AI workflow:

• create-prd.md — prompts the AI to produce a clear, actionable Product Requirement Document describing the feature, acceptance criteria, UX, edge cases, and relevant implementation notes.
• generate-tasks.md — converts the PRD into a list of small, ordered, atomic tasks (and sub-tasks) that can be executed one at a time by the AI.
• process-task-list.md (or analogous guidance) — instructs the AI to work through the task list one item at a time, run tests, and mark tasks complete.

These files are simple markdown templates with example prompts and usage instructions so you can call them inside your AI-enabled editor or CLI (e.g., “Use @create-prd.md” or tag a PRD file when generating tasks).

How it works (workflow)

1. Create a feature description and invoke create-prd.md to generate a formal PRD.
2. Feed the PRD into generate-tasks.md to produce a sequenced task list of small changes (1.1, 1.2, 2.1, …).
3. Ask the AI to start on the first task. The AI implements the change, runs tests (if available), and returns a patch or code suggestion.
4. Review, run tests locally, and approve or request changes.
5. Mark task complete and proceed to the next item. Repeat until the PRD is satisfied.

This linearization forces small-scope changes and frequent verification, dramatically improving reliability compared with large, undifferentiated prompts.

Practical usage examples

Example prompt to generate a PRD:

Use @create-prd.md  
Here's the feature I want to build: [Describe feature]  
Reference these files to help you: @app/models/user.py @app/routes/auth.py  

Generate tasks from a PRD:

Now take @MyFeature-PRD.md and create tasks using @generate-tasks.md  

Start working on a task:

Please start on task 1.1 from the generated task list. Implement the change, run unit tests, and provide the diff.  

Supported tools / integrations

• Designed to be tool-agnostic: works with Cursor, Amp, Claude Code, Windsurf, GitHub Copilot-like assistants, and other AI coding environments that support referencing local files and prompts.
• Works well in AI-enabled IDEs where you can tag or reference files (e.g., @MyFeature-PRD.md) so the assistant has context.

Benefits / “Superpowers”

• Forces discipline: clear PRDs + atomic tasks reduce ambiguity.
• Incremental verification: small changes + tests reduce regressions and make outputs reviewable.
• Solo-founder friendly: enables one person to scope, implement, and validate features that normally require a team.
• Portable templates: markdown files are easy to adapt to different languages, stacks, and coding assistants.

Limitations & caveats

• Not a silver bullet: quality still depends on the developer’s review, test coverage, and the AI model’s capabilities.
• Requires investment up-front in writing a good PRD and sometimes additional task splitting for complex features.
• Tooling limitations: some AI assistants may not fully support referencing local repo files or running tests automatically.

Pricing

• The repository itself is open-source and free to use. Costs come from the AI tools you use (paid API tokens, subscriptions to AI IDEs, compute, etc.).

References & resources

• Repository: https://github.com/snarktank/ai-dev-tasks
• Demo video referenced in README: Claire Vo “How I AI” podcast demo (linked from the repo README)
• Relevant forks / extensions: community forks that adapt the templates for specific stacks (e.g., PostgreSQL examples)

---