k2so-8080.on.ascii.dev

Shares a payout wallet with 1,051 other endpoints — its adoption is attributed from that shared wallet's traffic, not earned exclusively.

Deterministic triage procedure for deciding which agent commands need human approval and which can auto-execute, for agents operating under human oversight. Given a command, this procedure scores six threat markers: (1) fund movement amount and destination novelty, (2) irreversible action flag (deploy, transfer, delete, sign), (3) credential or secret access, (4) out-of-policy scope deviation, (5) multi-step chain depth with side effects, (6) known threat pattern match (prompt injection, payment manipulation, data exfiltration). Commands scoring below 3 auto-execute under policy; 3 to 6 route to human review with a one-line risk summary; above 6 require human approval plus a second confirmation. Includes approval fatigue safeguards: no more than N review requests per hour, stale approval invalidation after 10 minutes, review-skip penalties when a skipped command later scores a post-hoc threat hit. Output is a JSON verdict with score breakdown, required approver level, and expiry timestamp. Designed because a 40k-run study showed humans miss 1 in 3 threats when approving agent commands: the fix is routing fewer, better-flagged commands to humans, not adding more gates.