#
faq
security-war-room · knowledgebase & setup
GitHub
Documentation & Technical Answers
Questions, answered honestly
No marketing qualifiers. How reachability is evaluated, where analysis limits exist, what human guardrails apply, and how local agents execute on Switch.
It searches your source for imports and call sites of the function or setting the advisory names, and posts the file and line as evidence. It is a code search plus the agent's reading of it, not a full program analysis.
Calls made through dynamic names, and vulnerable code reached only through a transitive dependency. Then the analyst says "cannot confirm" and asks a human.
@patch-crafter tries a fix up to 3 times. If tests still fail it stops and posts the output. It never merges.
No. It opens a draft pull request and asks a named person for sign-off in the thread.
Any coding agent Switch supports (Claude Code, Codex, OpenCode) in Slack, Teams, Discord or Mattermost. The demo runs on the local Mattermost that ships with Switch.
Switch and the agents run on your machine. Your chosen coding agent sends code context to its own model provider, as it normally would.
The demo covers JavaScript projects with npm.
Get started
Bring the agents you already use. Clone the repo and run your first alert in about ten minutes.
$
git clone https://github.com/roshni09mishra-0/zeroday-sentry.git
about ten minutes to your first alert