You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
SDKArea: what it touches
Bug FixesKind: in v2.1.295,
Bug FixesSection of the release
What
The Agent SDK is the library that lets you run Claude Code from your own code. It starts the claude program with a list of options, called flags. It now passes many of those flags with the value joined on, such as --thinking=adaptive, instead of as two separate pieces. For flags including --model, --max-turns, --effort, --mcp-config and --managed-settings:
A value longer than one character that starts with - is joined to its flag, as in --max-turns=-1.
Any other value is still passed as a separate piece.
An empty --tools list is now passed as --tools= instead of --tools ''.
Why
A value such as a negative number or a string starting with a dash could be mistaken for a separate flag. That can no longer happen. If you wrap or inspect the claude command the SDK runs, expect both spellings.
--toolsClaude Code changelogmodified, high confidence* Fixed Claude being told to continue or message subagents with `SendMessage` in sessions, including resumed ones, where a host, a permission rule or a `--tools` list removed that toolsee the edit
--flagRun Claude Code behind a corporate launchermodified, medium confidence* **Don't depend on how arguments are spelled.** A flag's value can arrive as its own argument, `--flag value`, or joined to the flag, `--flag=value`. Which form a flag uses can change between versions.see the edit
How sure we are
Two sources agreeTwo things we can check say the same as this entry.
Anthropic's documentation agrees--tools on Claude Code changelog
Anthropic's release notes agreeFixed --tools and --restricted not applying to built-in tools that register after launch, and deprecated tool names reaching tools outside…
