Surface monthly AI usage limit errors across Studio Code surfaces

[youknowriad]

Related issues

How AI was used in this PR

Investigated, implemented, and tested with Claude Code, including live simulation of the 429 responses against a local fake proxy. Reviewed by the author.

Proposed Changes

When a user hits their monthly AI usage limit, the Studio Code agent silently stopped replying: the WordPress.com proxy's 429 was classified as a transient error and retried with backoff, and the final failure never reached the UI — the spinner just stopped with no message.

This PR makes hitting the limit fail fast and communicate clearly, consistently across every surface:

• No pointless retries: a 429 from the WordPress.com proxy (always the monthly cap there) now fails the turn immediately instead of burning ~15s of doomed retries. 429s on a user-supplied Anthropic API key keep retrying, since those are genuine transient rate limits.
• Errors are part of the conversation: a failed turn now renders an in-flow marker in the transcript — like the existing "Interrupted by you" marker — in the desktop app, the desktop agentic UI, and studio ui. The error is persisted on the session's turn_closed entry, so the marker survives reload. Non-cap errors show their actual message instead of nothing.
• Actionable, consistent copy: all surfaces share the same wording — "You've reached your monthly AI usage limit. It resets on <date>." — with the reset date fetched from the account's quota (the studio ui server gained a small quota proxy route for this). The CLI shows the same notice inline.

Testing Instructions

1. npm run cli:build
2. Start a fake proxy that returns the cap response:

   node -e 'require("http").createServer((q,s)=>{s.writeHead(429,{"Content-Type":"application/json"});s.end(JSON.stringify({type:"error",error:{type:"rate_limit_error",message:"Number of requests has exceeded your monthly limit."}}))}).listen(8931)'

3. Point any surface at it with WPCOM_AI_PROXY_BASE_URL=http://127.0.0.1:8931:
   • Desktop: WPCOM_AI_PROXY_BASE_URL=http://127.0.0.1:8931 npm start, open a site → Studio Code, send a message.
   • Local UI: npm run build:local -w @studio/ui, then WPCOM_AI_PROXY_BASE_URL=http://127.0.0.1:8931 node --experimental-wasm-jspi apps/cli/dist/cli/main.mjs ui, send a message.
   • CLI: WPCOM_AI_PROXY_BASE_URL=http://127.0.0.1:8931 node --experimental-wasm-jspi apps/cli/dist/cli/main.mjs code, send a message.
4. Verify the turn fails immediately (single request, no retry stall) and the usage-limit notice appears in the conversation flow with your account's reset date. Reload the session and verify the marker is still there.

⚠️ Visual change: the in-flow error marker needs human review in light + dark mode (verified in the browser UI in both schemes; the desktop renderer uses the equivalent --color-frame-error token).

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?

🤖 Generated with Claude Code

[wpmobilebot]

📊 Performance Test Results

Comparing 77194d8 vs trunk

app-size

Metric	trunk	77194d8	Diff	Change
App Size (Mac)	1362.04 MB	1362.06 MB	+0.02 MB	⚪ 0.0%

site-editor

Metric	trunk	77194d8	Diff	Change
load	1080 ms	1072 ms	8 ms	⚪ 0.0%

site-startup

Metric	trunk	77194d8	Diff	Change
siteCreation	7000 ms	7021 ms	+21 ms	⚪ 0.0%
siteStartup	2354 ms	2349 ms	5 ms	⚪ 0.0%

---

Results are median values from multiple test runs.

Legend: 🟢 Improvement (faster) | 🔴 Regression (slower) | ⚪ No change (<50ms diff)

[sejas]

Tested on both UIs and CLI. It works as expected. Thanks for adding a handler for the cost cap error.

Classic	Agentic	CLI