Building an Agent Tracking System with Claude
DetectiveMindless652 · reddit · 2026-07-15
The author used Claude to solve what they argue is the biggest pain point with AI agents: not memory or speed, but trackability, directional sense, and the costs of random erratic behavior.
After spending six months building a system, its core capabilities include:
- Identifying up to 6 types of loops, with email notifications to pause writes
- Categorizing and timestamping every agent action for easy traceability
- Shared memory among multiple agents (e.g., allowing billing and pricing agents to sense each other's changes)
- Built-in loop detection and cost functions to predict task expenses and potential loss points
- Auto-recovery for broken agents, along with failure cause localization
They also claim their from-scratch memory system outperforms mem0 on long evals, calling it one of the most advanced loop detection systems available. The project offers both cloud and local versions.
More from coding & agent
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22