Weekly Challenge

A new challenge ships every week. Each one is scoped to a single session — 1–3 hours of focused, buildable work. Fork a project, complete the challenge, share your result.

How it works

1. Read the current challenge below
2. Fork the repo or pull the relevant project
3. Build the solution in your own branch
4. Share on LinkedIn or GitHub Discussions with #AgenticPlaybook
5. The best implementations get highlighted in the following week's update

Current challenge

Week of 2026-05-12: Add streaming token counts to P01

Project: P01 — AI Provider Gateway

Goal: Extend the provider gateway to track and emit token usage during streaming responses. The final output should include prompt_tokens, completion_tokens, and total_tokens as a summary event at the end of each stream.

Acceptance criteria:

• The gateway accumulates token usage across streaming chunks where the provider supports it (OpenAI and Anthropic both surface this)
• A final usage summary event is emitted after the last content chunk
• The existing smoke test still passes: npm run smoke
• Add one test in examples/ that prints the usage summary

Hints:

• OpenAI streams usage in the final [DONE] chunk when stream_options: { include_usage: true } is set
• Anthropic emits message_delta and message_stop events with usage data

Share your result: Open a Discussion with your branch link and a brief description of what you built.

---

Past challenges

Week	Challenge	Project	Skills
2026-05-05	Add retry with exponential back-off to P01	P01 Gateway	resilience, error handling
2026-04-28	Add metadata filters to P02 retrieval pipeline	P02 RAG Copilot	vector search, filtering
2026-04-21	Add a human-in-the-loop approval node to P03	P03 Orchestrator	LangGraph, approvals
2026-04-14	Write an MCP resource for reading GitHub issues	P04 MCP Toolkit	MCP protocol, tools
2026-04-07	Add a live typing indicator to P05 Angular copilot	P05 Agentic UI	Angular, AG-UI streams

---

Propose a challenge

Have an idea for a good weekly challenge? Open an issue with the label weekly-challenge. Good challenges are:

• scoped to 1–3 hours of real work
• tied to a specific project in this playbook
• teachable — completing it produces a skill, not just a PR
• verifiable — there is a clear acceptance criterion

---

Challenge leaderboard

Consistent contributors who complete 4+ challenges and share their work publicly get a shoutout in the README and contributor wall. No points system — just recognition for people doing real work.