Review LLM and agent code: practice pull requests
Code that calls a model brings its own review questions: tool calls, streaming responses, context and token budgets, retries and rate limits, and where the API key comes from. These PRs wire a model into an app. Most hide a bug, some are clean, and none tells you which.
As of October 2026, the DiffDojo library has 15 live PRs in LLM and agent code, written in Python.
- Add a prompt-suite eval harness for the support pipeline
- Add a support chat endpoint grounded in help-center articles
- Add an AI refund agent to the support chat
- Add parallel tool execution to the support assistant
- Add per-conversation memory to the team chat assistant
- Add the BYOK chat proxy endpoint
- Add the LLM ticket-triage classifier endpoint
- Add the meeting-notes summariser endpoint
- Add the overnight ticket summariser with provider-aware backoff
- Nimbus assistant: chat endpoint with weather and calendar tools
- Stream the help-centre assistant's replies over SSE with a token-trimmed history
- Stream the support-chat answer to the widget over SSE
- Turn the invoice-note summariser into a batch endpoint with per-row outcomes
- Wire the get_weather tool into the weather assistant with idempotent execution
- Wire the order-lookup tool into the support assistant
Practise
Review an LLM PR with the answer hidden.
Leave inline comments and a verdict, then see what you caught. Free, no signup.
Start with the first one →