Tuesday, May 26, 2026

TSRX: TypeScript => UI for AI

 TSRX | A TypeScript language extension for building declarative UIs in an agentic era.

TSRX: A TypeScript Language Extension for Declarative UIs — A fresh attempt at improving upon JSX from a Svelte maintainer and former React core engineer.
It includes control flow, scoped styles, and locals, and compiles to React, Preact, Solid and 
Ripple.


JavaScript Weekly Issue 783: April 28, 2026



AI: tips to avoid Claude's limits

 21 tips to stop hitting Claude's limit: Post | LinkedInPost | LinkedIn

Sivasankar Natarajan | LinkedIn

1. Convert files before uploading
• Don't upload PDFs, screenshots, or PPTs. Copy the text into a Google Doc, download as .md, then upload.

2. Plan in Chat, build in Cowork
• Plan structure in Chat first. Then move to Cowork to build the actual file.

3. Ask questions
• Use clear prompts: "I want to [task] to [success criteria]." Ask Claude to "AskUserQuestion" before starting.

4. Stop redoing the whole thing
• If something is wrong, only redo the broken section. Add "no commentary, just the output."

5. Edit your original message
• If Claude responds poorly, click Edit on your original message, fix it, regenerate. Chat only.

6. Reuse the same prompt structure
• Build a library of prompts. Swap only the variables.

7. Batch tasks into one message
• Combine: "Summarize, list key points, suggest a headline" beats three separate prompts.

8. Pick the right model
• Sonnet or Haiku for grammar. Opus + Extended Thinking for real work.

9. Keep your files short
• Trim Cowork folders to under 2,000 words to avoid burning tokens.

10. Restart, don't follow up
• If something goes wrong in Cowork, click "Restart conversation from here".

11. Summarize every 15–20 messages
• Ask Claude to summarize key points. Copy the summary, start a new session, paste it as message one.

12. Don't dump your whole folder
• Only include files Claude needs for the specific task.

13. Use Projects for recurring files
• Stop uploading the same PDF every time. Upload once into a Project.

14. New topic = new chat
• Shifting topics? Start a new chat. Don't drag old context into unrelated work.

15. Turn off features you don't need
• Disable extra web search, connectors, and Explore modes you are not using.

16. Schedule recurring tasks
• Use the /schedule plugin to automate weekly digests.

17. Stop using Claude for things it can't do
• No image generation. No real-time search. Use Grok or another tool for those.

18. Speak your prompts for more context
• Use tools like Wispr to dictate prompts. More context, less reloading.

19. Set up preferences
• Set your style and turn off memory if not needed.

20. Prompt Claude Code tightly
• "Build a bar chart from this CSV. Save as chart.png." Tight instructions, fewer back-and-forths.

21. Spread across the day
• Claude works best in short sessions. Work in the morning, take a break, resume in the afternoon.

The takeaway

Limits are a budget. Heavy users do not get more tokens. They use the ones they have better. 
Tighter prompts, smaller files, the right model, and a habit of restarting.