agent-observability

Stream Claude Code agent events to a filesystem-based observability dashboard.

2|1|Updated Oct 11, 2025
One-click install
npx skills add https://github.com/rafaelcalleja/claude-market-place --skill agent-observability-rafaelcalleja
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-observability
Source: https://github.com/rafaelcalleja/claude-market-place/tree/main/plugins/personal-ai-infrastructure/skills/agent-observability
Command: npx skills add https://github.com/rafaelcalleja/claude-market-place --skill agent-observability-rafaelcalleja

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Understanding the complex, multi-agent interactions and decision-making processes of your AI can be opaque. This skill provides real-time observability to demystify agent activity, helping you debug and optimize workflows.

Core Features & Use Cases

  • Real-time Event Stream: Captures and displays all agent actions, thoughts, and tool calls as they happen.
  • Swimlane Visualization: Offers a clear, chronological view of individual agent contributions in multi-agent workflows.
  • Human-in-the-Loop (HITL) Notifications: Alerts you to critical events, allowing for timely intervention and course correction.
  • Historical Analysis: Stores and visualizes past agent interactions for post-mortem analysis and optimization.
  • Use Case: Debug why a multi-agent development task failed by reviewing the exact sequence of tool calls and internal thoughts of each agent, identifying bottlenecks or errors in real-time or historically.

Quick Start

Use the agent-observability skill to start monitoring all active AI agent events in real-time.

Frequently Asked Questions about agent-observability

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I monitor AI agent activity in real-time?

Real-time monitoring captures all agent actions, thoughts, and tool calls as they happen through event streaming. Agent-observability connects to Claude Code agent events, displaying them live in a filesystem-based dashboard without requiring a database.

Can I debug multi-agent workflows by viewing individual agent contributions?

Swimlane visualization organizes agent actions chronologically by individual agent, letting you trace exact sequences of tool calls, thoughts, and decisions across multi-agent workflows to identify bottlenecks or failures.

How do I set up observability for my AI agents without a database?

Agent-observability uses a filesystem-first architecture storing daily JSONL logs in ~/.claude/history/raw-outputs with an in-memory last-1000-events cache and WebSocket streaming, requiring no external database setup.

What happens when agents encounter errors—can I get alerts?

Human-in-the-loop notifications alert you to critical events including errors across multiple agents, enabling real-time intervention and course correction before workflows complete.

Can I review past agent interactions for analysis?

Historical analysis stores and visualizes past agent interactions through daily JSONL logs, allowing post-mortem debugging and optimization of multi-agent workflows.

Does agent observability work with tool usage and session tracking?

Event capture applies across tool usage, sessions, subagent launches, errors, and prompts, giving complete visibility into all agent activity types within a single observability dashboard.