Anthropic adds experimental Code Mode to Claude Code 2.1.323 for JavaScript tool chaining (product news, no vulnerability)
Anthropic has introduced an experimental Code Mode in Claude Code version 2.1.323 and later. It lets the assistant write a self-contained JavaScript program that runs locally in the session. The script can run tool calls in parallel, chain their outputs and filter the results, which reduces intermediate output and token use. The feature is disabled by default. It is enabled with the environment variable CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1, and the first launch asks the operator to confirm experimental execution privileges. The article reports no vulnerability, exploitation or threat activity. It is product news, and the only security-relevant point is that model-authored scripts run locally after the user opts in.
Detalhes técnicos
- Requires a native install of Claude Code 2.1.323 or newer, in an interactive terminal session.
- Requires authentication through a Claude subscription or Claude Console account. Third-party API relays and proxies are not supported.
- Supported models at launch are named in the article as Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.2.
- Ctrl+O shows the session transcript and the generated JavaScript.
- Not supported: VS Code panels, desktop apps, cloud sessions, the Agent SDK and non-interactive -p mode.
This record is based on the article alone. No web research was done, so Anthropic's own documentation was not checked.
Linha do tempo
-
Anthropic adds experimental Code Mode to Claude Code 2.1.323 for JavaScript tool chaining (product news, no vulnerability)
ler a análise desta fonte ocultar
Anthropic has introduced an experimental Code Mode in Claude Code version 2.1.323 and later. It lets the assistant write a self-contained JavaScript program that runs locally in the session. The script can run tool calls in parallel, chain their outputs and filter the results, which reduces intermediate output and token use. The feature is disabled by default. It is enabled with the environment variable CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1, and the first launch asks the operator to confirm experimental execution privileges. The article reports no vulnerability, exploitation or threat activity. It is product news, and the only security-relevant point is that model-authored scripts run locally after the user opts in.
Technical Details
- Requires a native install of Claude Code 2.1.323 or newer, in an interactive terminal session.
- Requires authentication through a Claude subscription or Claude Console account. Third-party API relays and proxies are not supported.
- Supported models at launch are named in the article as Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.2.
- Ctrl+O shows the session transcript and the generated JavaScript.
- Not supported: VS Code panels, desktop apps, cloud sessions, the Agent SDK and non-interactive -p mode.
This record is based on the article alone. No web research was done, so Anthropic's own documentation was not checked.