Back to Docs
Release History
Changelog
What's new in DeepSweep agent authorization
184 releases
77 security patterns
Latest: v2.0.98
Added
- The policy export now lists what the review found: each agent setting with its file and line, and each credential by its kind and where it sits.
Fixed
- A finding in a test file now says so at the start of its row.
Added
- An allow rule that lets Claude Code run any Python, Node or shell code without asking you is now reported in the Agent Environment Review.
Fixed
- Reviews read files faster, most of all on Windows, so large projects are less likely to run out of time.
- A review no longer stalls on a file with one very long line.
- A review no longer stalls on a file written to slow it down.
- Ordinary sentences that begin with words like Update or Delete are no longer reported as SQL injection.
- Code examples in README and other documents are no longer reported as problems in your code.
Changed
- A remote server with no sign-in in its settings is now rated medium instead of high.
- Ordinary comments and file paths in agent instruction files are now listed as medium instead of high.
- The notice that a new version is available now offers the download directly in Cursor.
Fixed
- Reviews of large projects now finish. A review reads every file instead of stopping after the first 200, and reads them faster.
- A review keeps going when files change. Editing or saving a file re-checks that file instead of restarting the whole review and clearing its results.
- When a review runs out of time on a large or slow project, your agent settings files have already been checked, and continuing the review to the end now finishes it.
- Fewer false findings. Hash values are no longer reported as Twilio credentials, the same CORS line is reported once instead of twice, and XML namespace addresses are no longer reported as plain-HTTP links.
- Four more kinds of false finding are gone: the word eval in ordinary text, letters inside encoded data read as an AWS key, placeholder values such as xxxxxxxx read as an API key, and quoted text ending in token= read as a secret.
Changed
- When one rule is found in more than five places, the findings list shows it as a single row with the number of places and files. Open the row to see each one.
- The findings list shows the most severe findings first.
- A finding in a test file now says so, on its row and in its detail, so sample credentials in tests are easy to tell apart. These findings still count toward your grade.