Debug Anything
Observe running processes, containers, and Kubernetes pods to generate evidence-backed diagnostics and profiling data for AI agents.
Debug Anything acts as a sensory runtime layer for coding agents, allowing them to diagnose live software issues without breakpoints, restarts, or code instrumentation. Compatible with Claude Code, Cursor, Codex, and Gemini CLI, it orchestrates standard diagnostic utilities like strace, py-spy, perf, lsof, and tcpdump across local PIDs, Docker containers, and Kubernetes pods. Rather than outputting raw logs, it structures observations into verifiable, ranked findings that cite specific evidence IDs, enabling autonomous agents to accurately pinpoint memory leaks, deadlocks, CPU spikes, and network bottlenecks.
Key Features
01Multi-target support spanning local host PIDs, Docker containers, and Kubernetes pods
0214 GitHub stars
03Zero-instrumentation profiling and diagnostics across native, Python, Node.js, Go, and Java runtimes
04Evidence-backed findings contract requiring agents to cite verified evidence IDs before drawing conclusions
05Rich artifact collection including CPU flamegraphs, thread backtraces, heap retainers, socket states, and pcap captures
06Symptom-driven automated analysis workflows for debugging leaks, hangs, crashes, and performance slowdowns
Use Cases
01Empowering AI coding agents to autonomously inspect and identify the root causes of production memory leaks and high CPU usage
02Diagnosing unresponsive microservices or deadlocked worker threads without stopping or restarting the target container
03Investigating networking issues such as unreleased socket descriptors, port collisions, or connection pileups