Repository navigation
Git worktrees, run limits and OpenAI cyber access checks #541
ezl-keygraph
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Shannon v3.6.1 is out. Together with v3.6.0, it brings three improvements that make scans easier to run in more setups.
Scan a git worktree
You can now point
-rat a git worktree, not just a regular clone:npx @keygraph/shannon start -u https://staging.example.com -r ~/src/my-app-feature-branchBefore, scans of a worktree couldn't use git inside the scan container. Now they work the same as a regular clone, so you can scan a feature branch from its own worktree without cloning the repository again.
Concurrency and timeout limits for slow model endpoints
If your model endpoint is slow or rate limited, such as a self-hosted model or a gateway with tight limits, two new settings help a scan finish:
SHANNON_MAX_CONCURRENCYSHANNON_AGENT_TIMEOUT_HOURSA lower concurrency makes the scan take longer, but the results are the same. Details are in the AI provider guide.
A more reliable OpenAI cyber access check
For OpenAI and OpenAI Codex models, Shannon now confirms with OpenAI directly whether your organization has cyber access before the scan starts. If access isn't enabled, the scan stops right away, instead of being refused partway through a long run. This works with an OpenAI API key and with a Pi login, including a ChatGPT Codex subscription. If you route OpenAI through a gateway with
SHANNON_AI_BASE_URL, make sure the gateway forwards OpenAI'saccess_programsrequest field.To check your account without running a scan:
If you haven't enrolled yet, see Anthropic and OpenAI Cyber Verification Programs.
Questions or problems? Reply here or open an issue.
All reactions