Repository navigation
An agent that logs in with stored credentials: what stops it doing more than the task? #8613
javmann100
started this conversation in
General
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.
Genuine question, not a hypothetical. page.agent.login with credentials from Skyvern, Bitwarden or 1Password, plus 2FA support, means the agent is acting as a real user on a real account. When a workflow asks it to do something ten times larger than usual (a bigger refund, a bulk change, a different account), what stops it today: the prompt, a limit in the workflow, or noticing afterwards?
I am building mnki, the fourth answer: bounded authority checked outside the model before an action runs (this agent, for this principal, these sites, these actions, these limits), a human approval on a phone for the cases that deserve one, and a signed record of every decision. It sits in front of a tool executor or an MCP server, so it fits alongside Skyvern rather than inside it.
Twenty minutes to hear how your customers handle it would help me more than any demo; a reply here is plenty. Integration paths: https://mnki.com/docs/integrations. Thank you for Skyvern; resistance to layout changes is what makes it usable on sites nobody has automated before.
All reactions