How modellismz/openskilltrace works
OpenSkillTrace is a governed AgentOps platform: operators build and preview workflows, then manage cases, approvals, audit, and operational fallbacks. Its flagship fraud-response flow collects a claim, gathers evidence, grounds decisions in RAG, classifies risk, and routes actions for human approval. The frontend talks to a FastAPI backend, which owns workflow execution, persistence, model routing, and integrations. MCP exposes backend operations to external clients; a separate worker processes queued RAG jobs. Some UI capabilities are README-documented but their implementation was not sampled.
Main entry points: main.py FastAPI backend · index.html Web application · rag.py Governed RAG · mcp_server.py MCP tools
Components 24
Operator interface
- Web application
index.html - Ops dashboard
assets/app.js - Workflow Studio
assets/studio.js - Cases and tickets
assets/views-notifications.js - War room
assets/views-warroom.js - Ops controls
assets/views-reliability.js
Workflow operations
- FastAPI backend
backend/main.py - Claim intake
backend/main.py - Evidence gathering
backend/main.py - Risk classification
backend/main.py - Preview runner
backend/main.py - Repair harness
backend/main.py - Human approval
backend/main.py - AgentOps JSON store
backend/main.py
Knowledge and governance
- Governed RAG
backend/rag.py - RAG job worker
backend/rag_worker.py - RAG SQLite index
backend/rag.py - Audit and policy
backend/main.py
External integrations
- MCP tools
backend/mcp_server.py - Model providers
- Qdrant
- Telegram Bot API
Other components
- Operations team
- MCP client
Connections 28
- Operations team → Web application: uses
- Web application → Ops dashboard: renders
- Web application → Workflow Studio: renders
- Web application → Cases and tickets: renders
- Web application → War room: renders
- Web application → Ops controls: renders
- Ops dashboard → FastAPI backend: calls API
- Workflow Studio → FastAPI backend: runs previews
- Cases and tickets → FastAPI backend: manages cases
- War room → FastAPI backend: calls API
- Ops controls → FastAPI backend: calls API
- FastAPI backend → Claim intake: handles claims
- FastAPI backend → Evidence gathering: gathers evidence
- FastAPI backend → Risk classification: classifies risk
- FastAPI backend → Preview runner: executes workflow
- FastAPI backend → Repair harness: creates repair
- FastAPI backend → Human approval: routes decisions
- FastAPI backend → AgentOps JSON store: reads and writes
- FastAPI backend → Governed RAG: retrieves knowledge
- FastAPI backend → Model providers: routes model calls
- FastAPI backend → Telegram Bot API: sends alerts
- RAG job worker → Governed RAG: processes jobs
- Governed RAG → RAG SQLite index: stores and searches
- Governed RAG → Qdrant: vector search
- FastAPI backend → Audit and policy: applies controls
- MCP client → MCP tools: invokes tools
- MCP tools → FastAPI backend: calls operations
- MCP tools → Governed RAG: calls RAG