simple10/agents-observe

What it solves

When AI agents (like Claude Code) run autonomously, their internal processes—such as spawning subagents, executing bash commands, and reading files—are often invisible to the user. This opacity makes it difficult to understand exactly what an agent is doing in real-time or to debug failures that occur deep within a parallel execution chain.

How it works

Agents Observe uses a system of hooks that trigger on every agent event (tool calls, prompts, and subagent lifecycles). A CLI tool captures these events and sends them via HTTP POST to a backend server (running in Docker) that stores them in a SQLite database. A React-based dashboard then streams these events in real-time via WebSockets, allowing users to visualize the agent hierarchy, tool usage, and session history.

Who it’s for

Developers using Claude Code or Codex agents who need deep visibility into the autonomous actions of their AI agents to debug, monitor, and analyze agentic workflows.

Highlights

• Real-time Monitoring: Watch tool calls and agent activity stream into a dashboard as they happen.
• Agent Hierarchy Visualization: See exactly which subagents were spawned by which parent agents.
• Token & Cost Tracking: Monitor token usage and cost breakdowns for each session.
• Full Event Replay: Browse historical sessions with full payloads, commands, and results for "time travel" debugging.
• Integrated Plugin: Can be installed directly as a Claude Code plugin with dedicated slash commands (e.g., /observe status) for management.