web fetch: https://example.com returns r.jina.ai's stale cached snapshot, not the page #1771

What happened

2026-10-05, dev@f81638787 (dev-20261005-f816387875e5). Found in the v0.7.1 happy-path pass.

$ codeaf web fetch https://example.com

It returns a page titled "Test Document" with a "test article" body, which is r.jina.ai's own stale cached snapshot and carries Jina's own "cached snapshot" warning. It is not the Example Domain page. codeaf web fetch https://www.iana.org/... returns the right page. The model's web tool goes through the same reader, so a model asked to read a page can be handed someone else's cache of it without knowing.

Replication

Field. OPENROUTER_API_KEY (the command needs a key), then run codeaf web fetch https://example.com. Compare with curl -s https://example.com.

Where

The Jina reader in internal/search/jina.go (and its use in internal/search/search.go), and the request headers it sends.

The fix

The fetch asks the reader for a fresh copy (Jina's no-cache header), or falls back to a direct GET when the reader marks its answer as a cached snapshot. What comes back is the page at that address now.

Acceptance

• e2e: codeaf web fetch https://example.com contains Example Domain.