AWS Permission Updates You Approve in episki, Never Beyond SecurityAudit
changelog·

AWS Permission Updates You Approve in episki, Never Beyond SecurityAudit

The AWS stack now registers itself, offers AWS's managed SecurityAudit policy, adds baseline posture checks, rolls out across an Organization, and can update itself when you click approve — with a permissions boundary that caps it at SecurityAudit. Plus dashboard trend history, one chat surface with edit and resend, agent sessions in the activity feed, and saved views on every list.

The AWS integration has always asked for a narrow, read-only role. The friction was everything around it: copying an account ID back into episki after launching the stack, and coming back to the CloudFormation console every time we added a check. This release removes both without asking you to trust us with more.

The stack introduces itself. On launch, a small function in the template tells episki the account ID, role, and stack version. The connect page is waiting for it and finishes the connection on its own. The function always reports success to CloudFormation, so episki being unreachable can never fail your stack.

You pick how permissions are maintained. SecurityAudit attaches AWS's own managed policy, which AWS keeps current, so new checks usually work without any update. Least privilege is episki's exact list and nothing else. The template defaults to least privilege, so an update made from the console or CLI never widens access unless someone chooses it.

Updates are one click, and bounded. If you opt in, an admin can approve a permission update from inside episki. The role episki updates has AWS's SecurityAudit policy as its permissions boundary, and the service role that applies the update can only change that one role's policies, only with that boundary in place. Whatever a template asks for, the effective access is metadata-only. Your existing parameters are carried over unchanged on every update.

The checks themselves grew too: CloudTrail, GuardDuty, Security Hub, Config, Access Analyzer, default EBS encryption, security groups, RDS, and KMS key rotation. A region we cannot read produces no result rather than a false "off", so a stack that has not been updated yet never generates findings it cannot back up.

Elsewhere, the dashboard now remembers. A nightly snapshot means the trend chart, the deltas, and the sparklines show where you were, not just where you are. And the agent's work reads like a conversation: one card per item, a clear state, and a place to reply when it needs input.

  • Self-registering AWS stacks and one-click, SecurityAudit-capped permission updates
  • AWS posture checks and an Organizations rollout with daily discovery
  • Dashboard trend history with a scrubbable chart
  • One chat surface with edit and resend, / skills, and real @-mentions
  • Agent sessions in the activity feed, and replies that instruct the agent
  • Saved views on every list, with Favorites in the sidebar

Share