Fix /news failing when Chromium binary is not installed

[konard]

Summary

Fixes #29 — /news reported every source as failed: browserType.launch: Executable doesn't exist at ... and prompted to run npx playwright install, collecting 0 items from all 51 sources.

Root cause

detectPlaywrightFetcher() decided whether to use the headless-browser fetcher or the plain-HTTP fallback like this:

const { chromium } = await import('playwright');
chromium.executablePath();          // ← only computes the path, never checks it
return new PlaywrightNewsPageFetcher();

chromium.executablePath() returns the expected location of the Chromium binary; it does not verify the file is actually present. On a machine where npx playwright install was never run (the user's Windows setup), it returns a non-empty path to a missing file, so detection wrongly picked the Playwright fetcher — and then every source failed at chromium.launch() time. The HTTP fallback that already existed in the codebase was never reached.

Fix

• Verify the binary exists on disk (existsSync(chromium.executablePath())) before selecting Playwright; otherwise use HttpNewsPageFetcher.
• Runtime safety net — new ResilientNewsPageFetcher wraps the Playwright fetcher and switches to the HTTP fallback the first time a browser-launch error occurs, so a partially-installed browser no longer fails every source.
• Clarified the README wording about when the HTTP fallback kicks in.

Tests

• ResilientNewsPageFetcher falls back to HTTP fetching on a launch error and does not retry the broken browser on subsequent calls.
• ResilientNewsPageFetcher rethrows non-launch errors (e.g. net::ERR_CONNECTION_REFUSED) instead of masking them.

All 128 tests pass; npm run typecheck and npm run build are clean.

How to reproduce / verify

On a machine without the Chromium binary installed, run /news. Before this change every source failed with Executable doesn't exist; after it, the crawler transparently uses the HTTP fetcher and collects items from static pages.

[konard]

Working session summary

The work is complete. Here's a summary of what I did.

Summary

Issue #29 — running /news failed every one of the 51 sources with browserType.launch: Executable doesn't exist at ... and collected 0 items, prompting the user to run npx playwright install.

Root cause

detectPlaywrightFetcher() in src/services/news-crawler.ts chose between the headless-browser fetcher and the plain-HTTP fallback by calling chromium.executablePath(). That call only computes the expected path of the Chromium binary — it does not verify the file exists. On a machine where npx playwright install was never run, it returns a non-empty path to a missing file, so the code wrongly selected the Playwright fetcher, which then failed at chromium.launch() for every source. The existing HTTP fallback was never reached.

Fix (src/services/news-crawler.ts)

1. Disk check before selecting Playwright — existsSync(chromium.executablePath()); if the binary isn't there, use HttpNewsPageFetcher.
2. Runtime safety net — new ResilientNewsPageFetcher wraps the Playwright fetcher and switches to the HTTP fallback the first time a browser-launch error occurs, so a partially-installed browser no longer fails every source. It rethrows non-launch errors so genuine failures aren't masked.
3. Clarified README wording about when the HTTP fallback activates.

Tests (tests/news-crawler.test.ts)

• Falls back to HTTP on a launch error and doesn't retry the broken browser.
• Rethrows non-launch errors (e.g. net::ERR_CONNECTION_REFUSED).

Verification

• All 128 tests pass; npm run typecheck and npm run build are clean.
• CI on PR Fix /news failing when Chromium binary is not installed #30 passed (commit 7148b61, conclusion: success).

PR: #30 (updated description, marked ready for review).

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $1.677268

📊 Context and tokens usage:

• 70.1K / 1M (7%) input tokens, 13.6K / 128K (11%) output tokens

Total: (2.9K new + 55.8K cache writes + 2.0M cache reads) input tokens, 13.6K output tokens, $1.677268 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.8 (claude-opus-4-8)

📎 Log file uploaded as Gist (1205KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag