Organize AI Development Tasks with Claude and GitHub Issues
Job to be done: Organize AI-driven development tasks using GitHub Issues as a database
🇳🇬 Ways to use this in Nigeria
Ideas to get you started, adapt to your situation.
- Entrepreneur
Building an AI customer-support chatbot for your store, use Claude to break the 'Automated FAQ Response' feature into tracked GitHub Issues like 'train on past queries' and 'integrate WhatsApp API'.
- Student
For your final-year AI inventory project, use Claude to break the 'Predictive Stocking' feature into tasks (data collection, model training, deployment), tracking each as its own GitHub Issue.
- 9-5 employee
As a fintech engineer building AI fraud detection, use Claude to split the 'Transaction Anomaly Detection' module into tasks (feature engineering, model selection, API endpoint), each a tracked GitHub Issue.
What you’ll get
A simple system for keeping AI-driven development organized by using GitHub Issues as your task database. AI agents lose track of context between tasks; this fixes that by writing every feature and sub-task down as a GitHub Issue, with Claude doing the planning and breakdown. You get a clear, traceable record of what needs doing and what is done, which matters even more when several AI agents work in parallel. It is intermediate: you use GitHub and run a few scripts.
Tools you need
- Claude (freemium): plans the project, writes the spec, and breaks features into tasks.
- GitHub Issues (free): your task “database”, each task is an issue, linked to its parent feature.
- bash (free): runs small scripts that create and link the issues automatically.
Steps
-
Plan the project with Claude: have it draft a Product Requirements Document (a “PRD”, a written description of what you are building and why):
Help me write a Product Requirements Document for an AI-driven [your project]. Focus on keeping context from getting lost between tasks and coordinating multiple AI agents. -
Create a GitHub repository: make a new repo to hold the project and its issues. You should see an empty repo on your dashboard.
-
Open an “epic” issue: create a GitHub Issue for a big feature (an “epic”, a large chunk of work that holds smaller tasks). Title it “Epic: [feature name]”.
-
Break the epic into tasks with Claude: feed it the PRD and the epic:
Based on the PRD and the epic "[feature name]", break the work into small, actionable tasks. For each, give a clear title and a one-line description. These become GitHub Issues. -
Create an issue per task: turn each task into a GitHub Issue, linked back to the parent epic (the author’s bash scripts automate this creating-and-linking; you can also do it by hand).
-
Track progress on the issues: as a task gets done (by you or an AI agent), update its issue, comment, change labels, or close it. The issues always show current status.
-
Keep it traceable: make every issue reference its spec or parent epic, so you can always trace a piece of work back to the requirement it came from.
Original source
Based on a Show HN post by aroussi, a lightweight project-management system using bash scripts and GitHub Issues to stop AI-driven development from losing context.
Notes & variations
- Free-tier viability: Claude’s free tier covers the planning, GitHub Issues are free, and bash is standard.
- Common pitfall: vague prompts produce vague tasks that are hard to track. Ask Claude for specific, single-purpose tasks.
- Tip for better results: use GitHub’s own features, labels for status, milestones for epics, to organize further, and let the bash scripts automate the repetitive issue creation.