// SYSTEM_SPEC
Agentic Workflow
Agentic Workflow demonstrates an autonomous agent execution loop with deterministic tool calling and a Human-in-the-Loop safety gate. Built using standard Next.js and FastAPI without heavy orchestration libraries.
- Category
- Autonomous Agents
System Overview
- Multi-step ReAct loop (Thought -> Action -> Observation -> Final Answer) with iteration guards.
- Deterministic local tools: safe AST math calculation, knowledge base lookup, and operational alerts.
- Human-in-the-Loop authorization: pauses execution on sensitive actions until user approval.
How to Use
- Select one of the suggested tasks or type a custom instruction.
- Observe the step-by-step reasoning and tool call trace in the execution timeline.
- If an alert is triggered, click Authorize or Reject to control agent execution.
Agent Capabilities & Guardrails
What the Agent CAN Do
- Safe Math Calculations: Evaluates arithmetic formulas, budgets, and cost projections via AST.
- Knowledge Base Queries: Searches internal policies (refunds, SLAs, on-call directory, cluster specs).
- Alerts & Notifications: Prepares operational warnings, emails, or chat broadcasts for channels/teams.
What the Agent CANNOT Do (Guardrails)
- Does NOT make phone calls or voice calls (no VoIP or telephony integration).
- Does NOT access arbitrary external APIs or open web beyond registered local tools.
- Does NOT execute critical alerts without consent: mandatory Human-in-the-Loop approval gate.
- Does NOT execute terminal commands or shell scripts.
Technologies Used
- Next.js 15
- React 19
- FastAPI
- Python 3.12
- ReAct Loop
- Human-in-the-Loop
- Tailwind CSS 4
- Pydantic 2