$ cat anthropics/week/2026-11-09.log
the week in review · Nov 9 – Nov 15, 2026
Claude Code 2.1.323 adds experimental Code Mode
By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology
Code Mode is opt-in and can be disabled per organization; v2.1.325 followed with a zstd decoding security fix and plugin version gating in plugin.json.
v2.1.323 anthropics/claude-code
by claude-code-publisher[bot]
Claude writes JavaScript that calls its tools instead of issuing tool calls directly; it is opt-in and can be turned off for an organization. The same release adds engines["claude-code"] to plugin.json, so a plugin declaring a Claude Code bound outside its stated range is gated at install rather than discovered at runtime.
v2.1.325 anthropics/claude-code
by claude-code-publisher[bot]
The release notes list a security fix in how zstd-compressed responses are decoded; no further detail on the failure mode is given. Also included: a one-time spinner tip about re-running a notebook cell after Claude reads or edits a Jupyter notebook, and an added line under organization errors.
Let maintainers say how reports and patches should look anthropics/oss-scanner
Maintainers can drop report_guidelines.md and patch_guidelines.md in their own repository at the path given in project.yaml, or next to it, to control the format of each report and the style of each patch. Both files are optional, and unlike threat_model.md they are read from the project rather than the scanner.
Pass structured output to shell steps through env anthropics/claude-code-action
GitHub interpolates a ${{ }} value into a run: script before the shell ever sees it, which is how structured output reached shell steps and where the injection path lived. Structured output now travels through environment variables instead, closing that path for action users.
fix(transport): fail closed on settings that can't be merged with sandbox (#1426) anthropics/claude-agent-sdk-python
When sandbox is set, the SDK merges options.settings into one JSON value, and two paths went wrong: a relative path resolved against the process working directory, and some settings could not be merged at all. The SDK now fails closed on both rather than proceeding with a partial merge.
fix(transport): look up claude.exe in absolute PATH entries on Windows anthropics/claude-agent-sdk-python
The Windows lookup took whatever shutil.which("claude") returned and patched around results it could not use: npm's claude.cmd shim, extensionless wrappers, claude.exe.cmd, relative paths. Only a native claude.exe is usable there, so the lookup now scans absolute PATH entries for one.
windows: caller env for MXC commands; SDK as optional peer anthropics/sandbox-runtime
spawnSandboxed() and MxcSandboxParams accept env, layered over every other source, and @microsoft/mxc-sdk with its koffi and node-pty native dependencies no longer installs for every user of the package. Embedders that need the MXC backend now supply it themselves.
nearcore: opt in to GitHub security advisories anthropics/oss-scanner
Alongside: Apache Doris enrolled branch-4.2-oss for scanning, crewjam-saml was enrolled, Yopass moved its Dockerfile and threat model into .oss-scanner/ in its own repository, nearcore opted in to GitHub security advisories, and security-guidance pinned its commit reviewer to the same model lookup as the diff review.
$ ls anthropics/week/ # the briefings behind this review
Keep up with Anthropic in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.
Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.