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

Commands named after JavaScript built-ins no longer confuse permission checks

You'll notice
Useful3 Signal3
Permissions

Commands literally named constructor or __proto__ no longer skew Bash permission checks.

What

Bash and PowerShell permission checking looked commands up in its tables by direct property access, so a command literally named constructor or __proto__ matched a built-in language property instead of missing. Lookups now go through a safe helper, so such names can no longer skew write detection or argument skipping.

Details
  • The affected tables are the ones listing command wrappers, options and verbs used to work out whether a command writes and which arguments to ignore.
  • A write with no determinable target still falls back to asking you: is a write operation but no target path could be determined; requires manual approval.
Evidence

is a write operation but no target path could be determined; requires manual approval

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 →