Source Intelligence

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

All of v2.1.246 Home All releases olderv2.1.245 v2.1.247newer
Claude Code v2.1.246

Auto-mode classifier prompt is now an embedded file, not JS source

Under the hood
Useful2 Signal3
Permissions

The permission classifier's instructions moved into an embedded file, so you can no longer read them from the JavaScript.

What

The full permission-classifier system prompt (the Environment, Definitions, HARD BLOCK, SOFT BLOCK and ALLOW document) was removed from the JavaScript bundle and is loaded from a text file embedded in the binary. The rules are unchanged, but the prompt can no longer be read out of the shipped JS.

Details
  • The module body is now a single load of /$bunfs/root/permissions_external-zcyf1abc.txt.
  • The loaded text is still scanned for the same four placeholder blocks covering user allow, soft deny and hard deny rules plus the environment, and spliced into the permissions template.
  • Unconditional; nothing gates the switch.
Evidence

/$bunfs/root/permissions_external-zcyf1abc.txt

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

See this entry in the whole of v2.1.246 →