Skip to main content
These admin routes run the Console’s own Trust Fabric audit reads for a platform API key. An External App is refused with 403, even when it holds the permission. Each read needs audit:read. The Tier 3 reads and the policy hold counts cover only decisions homed in the caller’s audit:read scope; the policies themselves have no home.
  • Tier 3 decisions: the recent free-form command decisions, each with its disposition, inferred risk, severity and the command’s hash. They come grouped by disposition, newest first within each, at most limit (default 200). The read takes twice limit Tier 3 decisions in that order before it applies the time window and your scope, so once that many share the first disposition, or the first disposition alone fills limit, decisions of the later dispositions are not returned. A decision never names the run, step or agent it was made for, and never carries its reason or gate messages.
  • Promotion candidates: frequent low-risk Tier 3 command patterns, by hash, count and last use, with no sample command line.
  • Policy effectiveness: each policy’s hold count, block rate and last trigger since a start time.
Three platform controls take switches:use with narrowed reach, each with a reason of at least 8 characters:
  • Engage the platform kill switch. Releasing it stays in the Console.
  • Set the publish judge to enforce. Moving it to warn or off stays in the Console.
  • Stop or resume the agents governance door. The people and api_keys doors stay in the Console.
A value outside that reach is 400, and nothing is written. Each control records the calling key as who acted.