Follow Discord
Sweep 25 Sep 2026 · 19:33Z Build v2.1.283 504 read Stable v2.1.274 Latest v2.1.283 Next v2.1.283 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.283 ·

Subagents limited in what they can do with account memory

Account memory servers can now set whether subagents get full, read-only or no access, and by default only the main agent can change memory

Group of 2 You'll notice Internal Changes
JSON All of v2.1.283
You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
4Signal: worth watching, 1 to 5
SessionsArea: what it touches
Internal ChangesKind: in v2.1.283,
What probably matters to youSection of the release

What

An account memory server is an MCP server, a plug-in tool provider, that stores things Claude remembers about you across sessions. Subagents are helper agents Claude starts to handle part of a task. They now have restricted access to that memory.

  • The server can advertise an experimental capability named anthropic/memory. Its subagent_memory field can be all, read_only or none, and subagent_note can hold up to 4096 characters.
    • If several servers set it, the most restrictive value wins.
    • A value that cannot be read counts as none.
    • If no server sets it, the mode is read_only.
  • The tools handed to a subagent are filtered by a new subagentMemoryMode. Tools from the memory server are removed unless they are on a read-only list, and none removes them all.
  • Forked slash commands and observer agents now pass the current permission mode as subagentMemoryMode when their tools are chosen.
  • A subagent, teammate, fork or backgrounded query that tries to change account memory is refused. The message says only the session's main agent can change it, and asks it to say in its reply what should be saved.
  • Subagents no longer see the memory server's resources, and reading one directly fails with a message pointing them to the server's tools.
  • Separately, --resume now records the source session's sourceSessionId and transcriptPath.

Why

Helper agents can no longer freely rewrite what Claude remembers about you. Changes to account memory go through the main agent, and the memory server decides how much subagents may read.

Read from
Names in the bundle--resume
What the documentation says
--resume Claude Code changelog modified, high confidence * Fixed more cases of continued or resumed sessions (`--continue`, `--resume`) re-sending earlier messages in a changed form, which could make the API drop Claude's earlier reasoning see the edit
--resume Changelog modified, high confidence * Changed session deletion to also remove the session's Claude Code transcript and related files from your computer, so `claude --resume` can no longer reopen it. see the edit
Since it was published

The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.

Confirmed since Anthropic's documentation has since written up --resume, on Hooks reference. `SessionStart` at launch, including with `--continue` or `--resume`, and every `Setup` event fire before the session's MCP servers are available to hooks. Claude Code skips their `mcp_tool` hooks without calling the tool, and the [debug lo… hooks see the edit
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 doubtIt is not clear what the new record or the passed permission mode changes for you in practice.
Anthropic's documentation agrees--resume on Claude Code changelog

See this entry in the whole of v2.1.283 →

Feedback