fix: --from-prompt timeout handling with actionable error output

[cosmicallycooked]

Bug

sonar interests create --from-prompt and sonar interests update --from-prompt call OpenAI/Anthropic with no request timeout. The OpenAI path uses the web_search_preview tool, which can legitimately take 30–60 s on a healthy connection. Any network hiccup, provider slowdown, or rate-limit event causes the spinner to hang indefinitely with no feedback — the process must be killed manually.

The same issue affects callOpenAIReply / callAnthropicReply (used in interactive mode).

Root cause

All four fetch() call sites in src/lib/ai.ts pass no signal, so requests have no deadline.

Fix

src/lib/ai.ts

Introduced fetchWithTimeout(url, init, timeoutMs, vendorLabel) — a small wrapper that attaches an AbortController to every AI request. On AbortError it throws a structured message that:

• Names the vendor and the elapsed timeout
• Lists three likely causes (provider load, network, web search latency)
• Suggests retrying or switching providers with --vendor

Per-vendor deadlines are generous to avoid false positives:

Vendor	Timeout	Reason
OpenAI	90 s	web_search_preview adds significant latency
Anthropic	60 s	Direct inference, no tool calls

Applied to all four call sites: callOpenAI, callAnthropic, callOpenAIReply, callAnthropicReply.

src/commands/interests/create.tsx + update.tsx

Spinner label for --from-prompt now includes the expected max wait time so operators can distinguish a slow-but-normal call from a genuine hang:

Generating interest via openai... (may take up to 90s with web search)

Verification

Reproduce by pointing the fetch at a server that stalls:

# Simulate a stalled OpenAI endpoint (not practical without mocking,
# but the timeout fires reliably under network isolation)
OPENAI_API_KEY=sk-test sonar interests create --from-prompt 'AI evals'
# Before: spinner runs forever
# After:  structured error after 90 s with vendor name, causes, and next steps

The timeout values (OPENAI_TIMEOUT_MS, ANTHROPIC_TIMEOUT_MS) are named constants in ai.ts for easy adjustment.

[Copilot]

Pull request overview

Adds request deadlines and clearer operator feedback for AI-backed --from-prompt flows so the CLI doesn’t hang indefinitely on slow/stalled vendor responses.

Changes:

• Introduces a fetchWithTimeout wrapper in src/lib/ai.ts and applies it to OpenAI/Anthropic interest generation + interactive reply generation calls.
• Adds per-vendor timeout constants (OpenAI 90s, Anthropic 60s).
• Updates the interests create/update --from-prompt spinner label to indicate expected maximum wait time.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 6 comments.

File	Description
src/lib/ai.ts	Adds a timeout wrapper around AI fetch() calls and applies it to all vendor request sites.
src/commands/interests/create.tsx	Updates the --from-prompt spinner label to include an expected max wait time.
src/commands/interests/update.tsx	Updates the --from-prompt spinner label to include an expected max wait time.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[cosmicallycooked]

@1a35e1 all threads resolved — ready for re-review.

Summary of changes pushed:

• fetchWithTimeout: Refactored to accept a processResponse callback. The AbortController timer now covers the full response cycle (headers + body). clearTimeout moved to finally for guaranteed cleanup.
• Timeout error message: Now vendor-aware — the web_search bullet only appears for OpenAI.
• Spinner labels (create.tsx + update.tsx): OPENAI_TIMEOUT_MS/ANTHROPIC_TIMEOUT_MS exported from ai.ts and imported here. getVendor() called once per render. Timeout values derived from the shared constants. 'with web search' qualifier is now conditional on vendor === 'openai'.