Claude Code 2.1.290 adds clearer agent hooks, sign-in controls and a long list of fixes

Claude Code 2.1.290 adds more detail to plugin hooks and subagent permission events, gives users a Deny button for pending sign-ins, and fixes a broad set of session, scheduling and access-control bugs. Anthropic’s release notes make this a substantial developer update, not merely a new number on the version label.

Anthropic Watch analysis

What happened

The release was published on 5 October. Hook authors can now see which tool calls an advisor ran, distinguish a subagent’s permission check from the main session’s, and see the approval an organisation requires. The release also adds validation output for gating hooks and colour types for plugin themes. Users can attach to a session or inspect its logs using part of its name, rather than needing the full session ID.

There are practical fixes across long-running and background work: scheduled tasks should resume more reliably after compaction or hand-offs, and resumed subagents retain their earlier thinking and prompt cache after receiving a message mid-run. WebFetch now reports when page text beyond 100,000 characters was unread and accepts an offset for reading further. The notes also describe fixes to permission checks and file access involving symlinks, including cases where image reads could reach outside an approved path. The Claude Code 2.1.290 release notes contain the full change list.

Why it matters

Claude Code sessions increasingly involve plugins, subagents and work continuing in the background. Better visibility into which agent triggered a permission check, plus clearer validation of hooks, gives developers and administrators more useful signals when something needs investigating. A Deny button that ends a pending sign-in also makes the choice concrete: approval need not be the only available action on the page.

The reliability fixes matter just as much. A scheduled task that quietly fails to return after a hand-off is not very helpful automation; neither is a tool that gives no indication it has stopped reading a long page. The release notes say these issues are fixed, but do not provide independent testing or quantify how often they occurred.

Our read

This is a worthwhile maintenance release with several changes developers can use immediately, especially those building plugins or juggling background agents. The permission and file-access fixes deserve attention, but the notes describe fixes, not a disclosed incident; no need to add drama where the release does not. If you rely on preview or plugin integrations, check the notes and test your own workflows after updating. Software remains impressively good at finding new ways to need a test run.

What to watch

• Whether plugin authors make use of the added hook and permission-event details.
• Whether the scheduled-task and subagent fixes hold up in long-running sessions.
• Whether subsequent releases clarify any migration or compatibility effects for plugin developers.

Discussion spark: For coding agents with plugins and subagents, which would improve your confidence more: clearer records of what each agent did, or stricter controls over what it can access?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.