Scans code for secrets, injection flaws and malicious patterns the moment you copy it. Works in AI chats. Runs locally.
Users
5
+4 in 30 days
Rating
0.0 / 5
unchanged in 30 days
Ratings
0
0 in 30 days
Store rank
#261K
#218K among extensions
Version
1.0.0
updated 7 Sep 2026
Users
monthly, 1 months
Latest, 16 Oct 2026:5 usersMove the mouse over the chart for any month
Ratings count
monthly, 1 months
Latest, 16 Oct 2026:0 ratingsMove the mouse over the chart for any month
The Chrome Web Store rounds user counts down to the leading digit, so the user line moves in steps. The ratings count is exact and is the better signal of real adoption.
About CodeGuardian Live
Copy a snippet from an AI assistant or a developer Q&A site, and CodeGuardian
reads it before you paste it. If it finds a hard-coded API key, a query built by
string concatenation, a shell command piped straight from the internet, or a
hash you should not be using for passwords, a panel names the problem, points at
the line, and shows the safer pattern.
It takes about twenty milliseconds, and nothing you copy ever leaves your
machine.
WHY IT RUNS WHERE IT DOES
Most scanners run in CI, which is after the mistake is already committed. This
one runs in the two seconds between copying and pasting, which is the only
moment the fix is still free.
It reads the structure of the page rather than guessing at it. On AI chat
surfaces the "Copy code" button dispatches no copy event at all, so a copy
listener alone is blind to the most common way people take code out of a chat;
that path is handled explicitly. And only the code in a reply is scanned, never
the prose around it, so an explanation of SQL injection is not reported as SQL
injection.
WHAT IT LOOKS FOR
83 rules in seven families:
- Secrets and keys, including a behavioural pass that catches credentials in
formats no catalog knows about
- Injection, with cross-line taint tracking from the request to the sink
- Weak cryptography, from broken hashes to disabled TLS verification
- Malicious patterns such as encoded payloads and remote scripts piped into a
shell
- Server-side request forgery, including cloud metadata endpoints
- Supply-chain risks hidden in install commands
- Deprecated and removed APIs
Thirteen programming and configuration languages are covered.
PRIVACY, ENFORCED IN CODE
- No network request exists anywhere in the extension. The content security
policy sets connect-src 'none', so one cannot be added by accident.
- No account, no telemetry, no server.
- The clipboardRead permission is never requested. The scanner reads the
selection inside the copy event itself, so it structurally cannot see
anything you copied somewhere else.
- History stores rule names and line numbers. Never the code you copied.
- Scanning is blocked outright on banking, authentication, government,
healthcare and password-manager sites, with no setting that can switch the
block off.
MEASURED, NOT CLAIMED
The test suites ship inside the extension folder: 128 engine assertions, full
recall on the benchmark corpus, 0.8 ms p95 scan latency, and no crashes across
3,000 fuzz cases. Run them yourself with node tests/run-tests.mjs
ALSO INCLUDED
- A searchable local history in the side panel, exportable as JSON
- A scanner for files you have just downloaded, read inside the tab and
never uploaded
- Ctrl+Shift+Y, or a right-click, to scan any selection on demand
- Pause on a single site without disabling the extension
- Light and dark, following your browser
It runs on the major developer Q&A, code-hosting, documentation and package
sites, and on the main AI assistants. The complete list is declared in the
extension manifest and published on the project page.