CodeMender Now Deep-Scans Your Repo by Default
CodeMender v0.16.0 landed on 21 October 2026. The headline is a changed default: cm find now runs a concurrent file-by-file deep scan unless you ask for the old behaviour with --quick. The newly scanned directories are the interesting part.
CodeMender v0.16.0, listed in Google Cloud's release notes for 21 October 2026, changes one thing that will show up in your next scan whether you read the notes or not. The release says cm find "now runs a concurrent file-by-file deep scan (--deep) by default," which raises vulnerability recall across a repository. The previous behaviour, a single-session quick scan, is still available behind a --quick flag.
A changed default is not a new feature. It is a new baseline for everyone who has that command in a script, a pre-commit hook, or a nightly job.
What the flip actually changes
A file-by-file deep scan reads each file as its own unit of work and runs them concurrently, rather than packing the repository into one session. The practical effects, in order of how soon you will notice them:
More findings, including ones that were previously missed because a large repository exhausted a single session's attention before reaching them.
More work per run, because concurrency raises throughput but the total analysis is larger.
cm statsnow breaks usage down by mode, across--deep,--quick, and--diff, so you can see which mode your spend went to.
If your pipeline has a time budget, the honest move is to measure before you let the new default run unattended on a large monorepo. If it has a cost budget, cm stats is now the place that tells you what happened.
The newly scanned paths are where the surprises live
The same release widens coverage in two ways that will produce findings in code people forgot was in the tree. On file types, cm find now scans .hh, .hxx, .inl and .ipp C++ headers, and it now reads extensionless scripts whose shebang points at Python, Node, Ruby, or PHP.
On directories, it now scans nested build/, target/, obj/ and bin/ source directories, plus production code sitting under spec/, locale/ and i18n/.
Those last three are the ones to think about. Plenty of projects keep real code in a locale or i18n directory, template helpers that interpolate strings, small formatting utilities, occasionally a fixture loader that reads from disk. It was outside the scan yesterday and is inside it today. Expect a first deep run on an older repository to return findings that are genuinely yours and simply were not being looked at.
A build output directory with checked-in generated sources is the other common case. If your first post-upgrade scan lights up in build/, check whether the finding is in generated code before you spend an afternoon on it.
Also in this release
A new cm health command checks credentials, configuration, and endpoint reachability before a scan runs, which is a small but welcome addition for anyone who has watched a CI job fail twenty minutes in on an expired token. The release also carries assorted bug fixes.
From research project to command line tool
Worth placing in context, because the name is older than the CLI. Google DeepMind introduced CodeMender on 6 October 2025 as an AI agent that finds the root cause of a vulnerability and writes a patch for it, using Gemini Deep Think models for the reasoning. At announcement it had upstreamed 72 security fixes to open-source projects over six months of development, and DeepMind was explicit that "all patches generated by CodeMender are reviewed by human researchers before they're submitted upstream."
That human review line is the one to carry forward into your own use. A tool that proposes patches is a tool that proposes patches. The flipped default raises how much it finds, not how much you can accept unread.
What to do this week
Grep your CI config and local scripts for cm find. Decide deliberately whether each one wants --deep or --quick, rather than inheriting the change.
Run one deep scan on your largest repository and time it before you put it in a hook.
Triage the first wave of findings by path. Anything new under
locale/,i18n/or a build directory is probably newly scanned rather than newly broken.Add
cm healthto the front of any scheduled job.
If the findings land in code an agent wrote for you, the reading order matters: our guide to reviewing AI-generated code before you ship it pairs with this, as does the prompting pattern for auditing your dependencies. And if a scan hands you an identifier you do not recognise, start with what a CVE is and is not.
Tracking releases like this one is the slow half of security work. Our note on keeping up with AI news without drowning in it covers how we pick which changelog entries are worth a post.
How did this land?
About the author

Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.


