Add Open in ChatGPT and Open in Claude page actions

[bilal-karim]

Summary

Adds two new external-link actions alongside the existing Copy for LLM and View as Markdown CTAs under the docs page title:

• Open in ChatGPT — opens chatgpt.com in a new tab with a prompt prefilled to read the current page
• Open in Claude — opens claude.ai in a new tab with the same prefilled prompt

The prompt sent to each assistant is: Read <page-url> and answer questions about the content.

Changes

• src/components/LLMActions/LLMActions.tsx
  • Capture the page URL client-side (window is unavailable during SSR) and build encoded ChatGPT/Claude deep links
  • Render the two new actions as anchors with the OpenAI/Claude brand icons (react-icons/si) plus a trailing external-link icon
  • Carry the same data-analytics-* attributes as the existing actions (open-in-chatgpt / open-in-claude) so usage is tracked in Amplitude
• src/components/LLMActions/LLMActions.module.css
  • Normalize height between <button> and <a> elements (line-height, box-sizing, font-family, text-decoration)
  • Add a muted, smaller externalIcon style

Testing

• Verified locally via npm run start; all four actions render at a consistent height under the page title and the external links open the respective assistants with the page URL prefilled.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
temporal-documentation	Ready	Preview, Comment	Jun 4, 2026 8:53pm

[github-actions]

📖 Docs PR preview links

This PR does not change any pages in /docs. If you make updates, links to the modified pages will appear here.

[alfredsmchan]

LGTM

[jsundai]

Nice!! Will take a full look sometime today.

[alfredsmchan]

if it fails to get the Page URL, we should hide the buttons unless there is a valid URL.

[bilal-karim]

Good call — fixed in 3e09dfc, though I went a slightly different route than disabling the buttons.

Root cause: the prompt URL was built from window.location.href captured in a client-side useEffect, so it was empty during SSR and the first render — that's the window where a click would produce Read and answer questions.... (Note: the "failed to fetch" you saw is actually a separate path — that's Copy for LLM fetching the raw markdown, not the page-URL resolution. window.location.href itself is synchronous and doesn't throw.)

Rather than guard/disable on an empty value, I removed the failure mode entirely: the URL is now derived synchronously from Docusaurus's own data —

const { siteConfig } = useDocusaurusContext();
const pageUrl = `${siteConfig.url}${metadata.permalink}`;

So it's a valid canonical URL during SSR, first paint, and every render — no useEffect, no state, no empty-URL window to defend against. Bonus: it's the clean canonical URL (e.g. https://docs.temporal.io/evaluate/) rather than whatever query/hash happened to be on window.location.

Verified in the browser — the rendered links resolve to:
Read https://docs.temporal.io/evaluate/ and answer questions about the content.

[alfredsmchan]

Thanks for update!