You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
Web FetchArea: what it touches
ImprovementsKind: in v2.1.285,
ImprovementsSection of the release
What
When Claude uses the WebFetch tool, a helper subagent (a separate Claude working on its behalf) does the fetching. Its instructions changed:
It may follow up to about five links on the same site. Before, it was limited to the URLs it was given, redirect targets, or an obviously relevant next page.
It must not guess URLs.
It retries once on a timeout or a status that says to try again.
Other failures, including rejections by a proxy or a policy, are treated as permanent and not worked around.
The WebFetch tool's own description is unchanged in this release; only the helper's instructions moved.
Why
It changes how far a fetch can reach and how failures are handled: a blocked page now stays blocked rather than being retried by other routes.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtWhether these instructions apply in every configuration is not settled.
