The Two Birds Prompt Library
Prompts built from real work, not demos. Each one solved a real problem and produced a real result.
Why It's A Eureka Moment
Most people think of AI assistants as interactive. This prompt turns Claude Code into an autonomous build agent that reads your own planning documents as its instructions. Your backlog file becomes the control plane. The log file becomes the accountability layer.
The Verbatim Prompt
You are an autonomous build agent for Two Birds Innovation. Read C:\twobirds\two-birds-portfolio\NEXT-SPRINT-QUEUE.md and C:\twobirds\two-birds-portfolio\SESSION-STATE.md. Execute the top 3 items that are fully Claude Code executable — skip anything marked as Aaron manual, pending approval, or requiring human input. Commit and push after each completed item. After completing all tasks, append a run entry to C:\twobirds\two-birds-portfolio\logs\automated-run-log.md with: date/time, tasks completed, commits made, anything skipped and why, next recommended action. Then commit and push the log file. Finally, overwrite C:\twobirds\two-birds-portfolio\logs\LAST-RUN-SUMMARY.md with a plain-text summary of what ran, then commit and push it. Canadian English. Static HTML/CSS/JS only. No npm. No Node frameworks.
How To Adapt It
Replace file paths with your own backlog and state files. Change "top 3 items" to however many you want per run. The key insight: your planning document IS the prompt input — keep it machine-readable.
Why It's A Eureka Moment
A raw GitHub URL is a universal, persistent, always-current API endpoint for your own work. You don't need a dashboard, a Slack bot, or a notification system. You just need a file that gets overwritten after every session.
The Verbatim Prompt
Finally, overwrite C:\twobirds\two-birds-portfolio\logs\LAST-RUN-SUMMARY.md with a plain-text summary of what ran, then commit and push it.
How To Adapt It
Create your own LAST-RUN-SUMMARY.md in any GitHub repo. Add the overwrite instruction to the end of every autonomous prompt. Bookmark the raw GitHub URL for quick access from any device.
Why It's A Eureka Moment
Most onboarding optimises for speed. For anxious users, speed is the enemy. They need reassurance before action. The insight: for trust-sensitive audiences, the onboarding IS the product experience.
The Verbatim Prompt
Build a first-visit onboarding overlay for the Digital Confidence Centre.
Target audience: seniors 70+ who are afraid of technology.
4 steps, each one screen:
Step 1: "Welcome to the Digital Confidence Centre" — what this is in one sentence
Step 2: "This is completely free and safe" — safety reassurance, nothing downloads,
you cannot break anything, your information stays private
Step 3: "Start with what matters to you" — show 3 module categories
(Staying Safe, Staying Connected, Getting Things Done)
Step 4: "You're ready" — single green button: "I'm ready to start"
Large text. High contrast. One action per screen. No jargon.
Remember on return visits (localStorage flag). Dismissible but not aggressive.
How To Adapt It
Change the audience profile — anxious users exist in healthcare, finance, and government. Replace the safety messages with whatever your users fear most. Keep the one-action-per-screen principle. Name the fears out loud.
Why It's A Eureka Moment
A markdown file committed to git IS a memory system. It's version-controlled, portable, human-readable, and machine-readable. The git history gives you time travel for free — you can see what Claude knew at any point in the past.
The Verbatim Prompt
Read C:\twobirds\two-birds-portfolio\SESSION-STATE.md before starting. [... do work ...] FINAL STEP — Create or update SESSION-STATE.md with: - Date and time of this session - Phases that ran and their completion status - Commits made (repo, commit message, push status) - Anything skipped and why - Next recommended action
How To Adapt It
Create a SESSION-STATE.md in your project root. Add "Read SESSION-STATE.md before starting" to the top of every prompt. Add the "FINAL STEP" template to the bottom. Keep the structure consistent. For multi-repo projects, keep state in one central repo.
Why It's A Eureka Moment
Most people use AI to do work. This prompt uses AI to watch you work. Every friction point is a potential product improvement, process fix, or automation opportunity. Your daily frustrations are data, and the pattern in that data is where the leverage lives.
The Verbatim Prompt
While working on this session, observe my workflow for friction points. A friction point is any moment where: - I repeat a manual step that could be automated - I hit an error that wastes time - I work around something broken instead of fixing it - I context-switch unnecessarily - A tool fails and I have to retry Log each friction point as: FP-[number]: [description] | Score: [1-10] | Status: LOGGED At the end of the session, write all friction points to a FRICTION-LOG.md file. Sort by score descending. For anything scored 7+, add a "Suggested fix" line.
How To Adapt It
Add this instruction to the top of any extended work session. Adjust the friction criteria. Review the log weekly — patterns across sessions reveal systemic issues. Share friction logs with your team to normalise talking about inefficiency.
Why It's A Eureka Moment
The guide isn't generated from documentation — it's generated from pain. Every step includes the workaround for the thing that actually breaks. These gotchas aren't in any official docs. They come from setting up two real machines and hitting every wall.
The Verbatim Prompt
Create a complete step-by-step guide for setting up Claude Code on a new Windows machine. Based on hard-won learnings from i5 Lenovo and silver Pentium setups. Include every section below exactly: - BEFORE YOU START: right-click paste, admin PowerShell, credentials needed - STEP 1: Install prerequisites (Git, GitHub CLI, Node.js via winget) - STEP 2: Install Claude Code (npm install -g @anthropic-ai/claude-code) - STEP 3: Authenticate GitHub and Claude - STEP 4: Create workspace and set permissions (icacls) - STEP 5: Clone all repos - STEP 6: Set git identity - STEP 7: Create desktop launcher - STEP 8: Disable sleep on AC power - STEP 9: Register overnight scheduler - STEP 10: Verify everything works - KNOWN GOTCHAS: every real error encountered and its fix
How To Adapt It
Replace the repo list with your own. Adjust the workspace path. Add or remove steps based on your stack. The gotchas section is the most valuable part — add your own as you discover them. A setup guide written from memory of what broke is worth ten times one written from docs.
Have a prompt worth sharing?
If you've built something with Claude Code that solved a real problem, we'd love to hear about it.
Suggest a prompt