AgentCert Workflow

OTEL • Langfuse • LiteLLM • MCP • MongoDB
SIMULATION PHASE — RUNS 1-30
0 / 30
simulations
select
inject
exp details
MCP
stores ←
pull ←
▲ Per-run pipeline: Inject → MCP Trace → Bucketing + Extraction × 30
▼ Post-30: Aggregation → Hypothesis Framework → Certification
1
Fault Library
Application
Network
Resource
2
Chaos Center
Inject faults + exp details → Langfuse
● container-kill
● pod-dns-error
● disk-fill
3
SockShop App
K8s • OTEL instrumented
front
cart
order
user
pay
ship
catal
DB
Langfuse
Trace & experiment store
↔ Chaos Center, ITOps, Certifier
4
ITOps Agent
SockShop ↔ ITOps via MCP
◆ MCP — K8s Trace Pull from SockShop
Langfuse Store
LiteLLM Proxy
Sidecar (exp_id)
5
Certifier
Per-run processing
BucketingExtraction
6
MongoDB
0 / 30
Certifier — Post 30 Runs
Final pipeline (MongoDB + Langfuse data)
7
Aggregation
8
Hypothesisoptional
9
Certification
Certificate
Report & JSON output
CERTIFIED
SockShop Agent Certification Report
ITOps Agent • Resilience & Safety Evaluation
30 Runs
3 Fault Categories
OTEL Traced
Langfuse Observed
Hypothesis Tested
Mitigation Rate
100%
Safety (RAI)
1.00
Reasoning
8.48
Detection
13.3%
Hallucination
0.00
Security
1.00
Fault Categories Tested
Application (10)
Network (10)
Resource (10)

Pipeline Steps

① Fault selected
② Injected + exp → Langfuse
③ SockShop OTEL traces
④ MCP: SockShop → ITOps
   LiteLLM + Sidecar
   ITOps → Langfuse
⑤ Langfuse → B+E
⑥ Stored in MongoDB

Current Fault

—
—

Experiment

exp_id: —
run_id: —
Initializing pipeline...