Follow Discord
Sweep 22 Sep 2026 · 17:19Z Build v2.1.280 501 read Stable v2.1.267 Latest v2.1.280 Next v2.1.280 Feeds RSS JSON llms.txt Unofficial
Claude Code v2.1.271 ·

Artifacts get 'path pins' to catch stale writes, plus an overwrite_unread escape hatch

Artifact publish now refuses to overwrite a file that changed since Claude last read it, unless explicitly told to with overwrite_unread

TierUnder the hoodhow much it should matter to you
Useful1my rating, 1 to 5
Signal2worth watching, 1 to 5
AreaArtifactswhat it touches
KindInternal Changesin v2.1.271,
Group of 5 Under the hood

Artifact publish now refuses to overwrite a file that changed since Claude last read it, unless explicitly told to with overwrite_unread

What

  • A new "path pins" system tracks the sha256 hash and version of every artifact file Claude has read or listed. Publishing (writing or removing) a file is now refused if the live file has changed since Claude last saw it, or if Claude never read it at all, with explicit "Not published"-style messages prompting Claude to re-read before retrying. Batched file reads are capped, with an exemption for live files.
  • This state (pathPins, pathPinAck, frozenPathPin, pathPinSeq) is cleared on session reset, unless the reset continues the same conversation.
  • A new overwrite_unread parameter on artifact publish (for both files and root publishes) lets a caller explicitly name files to replace or remove even though Claude hasn't read or listed them this session, for cases where the user explicitly asked for an outright replace.

Why This prevents Claude from accidentally clobbering artifact changes it hasn't seen, for example edits made by someone else since Claude last looked, while still allowing an explicit, deliberate overwrite when that is genuinely what the user wants.

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 doubtWhat a 'path pin' actually restricts or means for the user is not explained in the finding.

See this entry in the whole of v2.1.271 →

Feedback