Follow Discord
Sweep 28 Sep 2026 · 18:16Z Build v2.1.284 505 read Stable v2.1.277 Latest v2.1.284 Next v2.1.284 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.284 ·

Auto-memory and pinned memory are cleaned before use

Automatically written memory and pinned memory now have hidden characters cleaned out and tag-like text escaped before they reach Claude

Group of 2 You'll notice Improvements
JSON All of v2.1.284
You'll noticeTier: how much it should matter to you
1Useful: my rating, 1 to 5
2Signal: worth watching, 1 to 5
MemoryArea: what it touches
ImprovementsKind: in v2.1.284,
ImprovementsSection of the release

What

Auto-memory is the notes Claude Code writes for itself to remember things between sessions. Pinned memory is memory marked to always be included. Both are now cleaned before Claude sees them.

  • When memory files of type AutoMem and AutoMemPinned are loaded, line endings are normalised: \r\n, \u2028 and \u2029 all become \n. Control and formatting characters also appear to be removed. Before, only AutoMem got this treatment, and pinned content was used exactly as read.
  • If cleaning changes a file, Claude Code notes that the content differs from what is on disk and keeps the original text alongside it.
  • AutoMem entries and <pinned-memory> blocks are now escaped before they go into the prompt. A backslash is placed after the < of anything that looks like an opening tag, so memory text cannot open or close the tags Claude Code uses to structure the prompt. Before, the content was only trimmed.

Why

Memory written automatically can pick up hidden characters or text that looks like instructions or tags. Cleaning and escaping it keeps such content from changing how the rest of the prompt is read.

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 doubtExactly which characters are removed is not settled.

See this entry in the whole of v2.1.284 →

Feedback