View traces
Agent Detail
Soteria Idle
Risk & Compliance
85
Healthy
↓ -6 vs baseline
Per-agent score blending task success, exec time, error rate, and cost-per-task for Soteria.
Task Success 90% sub-score 90 / 100
Avg Exec Time 0s sub-score 100 / 100
Error Rate 0.0% sub-score 100 / 100
Cost / Task $0.32 sub-score 28 / 100
ROI Attribution
Fleet total this week — 94h saved · 847 tasks · ~$6,200 saved
Soteria
Time Saved 12h
Tasks Done 112
~$840 saved this week
Eval Gate Status
Pre-deploy safety net — shared fleet gates this agent must pass before going live.
These five automated gates run before any agent change reaches production. The status below reflects the most recent fleet-wide run — a new deploy for Soteria is blocked if any gate falls below its threshold.
Smoke Test
Schema & Output Happy-path output schema matches expected shape for every agent handler.
100% schema match · 500 tasks 2 min ago
Regression Suite
Golden Tasks Replay of 312 frozen golden tasks matches prior baseline within tolerance.
≥ 98% task-level parity 4 min ago
Cost Ceiling
Unit Economics Per-task token spend stays under the $0.42 ceiling tied to the Fleet Health widget.
p95 ≤ $0.42 / task 6 min ago
Compliance & Brand-Safety
Soteria Domain Matches the domain checks the Soteria agent already enforces — blocklist, sender reputation, regulated verticals.
0 hard flags · brand score ≥ 90 8 min ago
Red-Team / Adversarial
Safety Prompt-injection and jailbreak attempt set. One flaky category — agent handles partial leaks but not sustained extraction.
0 successful exfils · ≤ 2 partial leaks 11 min ago
Execution Traces
5 steps · most recent run on Soteria
1 Reputation check 2:00 PM ›
State at step 1
complete · 2:00 PM
Ran reputation check on 12 new domains in pipeline.
2 Audit sequences 3:50 PM ›
State at step 2
complete · 3:50 PM
Audit complete — 94% sender reputation score.
3 Update blocklist 4:30 PM ›
State at step 3
complete · 4:30 PM
Updated blocklist: 3 domains flagged, 1 removed.
4 Validate active sequences 5:15 PM ›
State at step 4
complete · 5:15 PM
Validated all active sequences — 0 compliance flags raised.
5 Scheduled next audit — ›
State at step 5
pending · —
Next compliance audit scheduled for 6:00 PM.
Memory Graph
7 context items · most-recent first
Soteria Hard-fail rule — regulated-… Hard-fail rule — sender rep… Never autoflush or override… 0 compliance flags across 1… Audit complete — 94% sender… Blocklist updated — 3 domai… Brand-safety ruleset rev 5c…
Version History
@soteria-prompt-history
5c014e8 Jul 10, 5:11 PM @alex 96 ↑ +5 
You are Soteria, the compliance agent.
Goal: screen every outbound message against the active blocklist and brand-safety rules.
Hard-fail on any regulated-vertical language (finance, health, legal claims).
Hard-fail if sender reputation < 80.
Emit a structured flag with category, severity, and the offending phrase.
Never autoflush or override a hard-fail.
9f6b302 Jun 22, 12:48 PM @mira 91 ↑ +5 
You are Soteria, the compliance agent.
Goal: screen outbound against the active blocklist.
Hard-fail on any regulated-vertical language.
Hard-fail if sender reputation < 80.
Emit a structured flag with the offending phrase.
1b8a47c Jun 5, 4:00 PM @jordan 86 — 
You are Soteria, the compliance agent.
Goal: screen outbound against the active blocklist.
Hard-fail on any regulated-vertical language.
Emit a structured flag with the offending phrase.