Follow Discord
Sweep 27 Oct 2026 · 19:04Z Build v2.1.312 531 read Stable v2.1.308 Latest v2.1.312 Next v2.1.312 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.312 ·

Groundwork for quoting tool input safely without leaking secrets

A helper that quotes a tool's input as JSON and refuses when it would expose a hidden secret is in the build, with nothing using it yet

Under the hood Internal Changes
JSON All of v2.1.312
Under the hoodTier: how much it should matter to you
1Useful: my rating, 1 to 5
2Signal: worth watching, 1 to 5
PermissionsArea: what it touches
Internal ChangesKind: in v2.1.312,
Internal ChangesSection of the release

Unclear Nothing is seen using this helper, so whether it affects anything a user sees is not known.

What

Claude Code now contains a helper that takes the input Claude gave a tool and tries to quote it as JSON, a common text format for structured data. A tool is one of the actions Claude can take, such as running a command or editing a file. The helper gives one of four answers:

  • quoted, when the input was quoted
  • too_long, when the input is too long to quote
  • not_json, when the input is not valid JSON
  • holds_secret, when the input holds a value that would normally be hidden and that value does not also appear in a second piece of text it is compared against

Nothing in Claude Code calls this helper yet.

Why

This looks like preparation for showing a tool's input inside other text, such as a transcript, while making sure a secret like a password or key is never copied out where it was not already visible.

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 doubtNothing is seen using this helper, so whether it affects anything a user sees is not known.

See this entry in the whole of v2.1.312 →

Feedback