Siddhant-K-code

Siddhant-K-code / agent-trace

Public

strace for AI agents. Capture and replay every tool call, prompt, and response from Claude Code, Cursor, or any MCP client

siddhantkhare.comwritingagent-observability-gap agent-observability agent-trace agent-traces agents ai-agents
14
2
100% credibility
Found Mar 22, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Agent-trace records and replays the complete interactions of AI coding agents to help users understand, debug, and optimize their behavior.

How It Works

1
😩 Frustrated with mysterious AI agent actions

Your AI coding helper changes files or runs tests, but you have no idea what steps it took or why it failed.

2
📥 Grab the tracing helper

Install this simple tool that watches your AI agent without any hassle.

3
⚙️ Link it to your AI coding tool

Paste a quick setup snippet into your tool's preferences so it starts recording quietly.

4
🤖 Work with your AI agent as usual

Give it tasks like fixing bugs or running tests – everything gets captured automatically in the background.

5
🔍 Replay the full adventure

Watch a colorful timeline replay showing every prompt, tool use, response, and mistake your agent made.

6
📊 Get smart insights

Review plain-English explanations, spot retries and waste, and estimate how much it cost.

🎉 Unlock AI agent mastery

Now you see exactly what your agent did, debug faster, cut waste, and make it way more reliable.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is agent-trace?

agent-trace brings strace to AI agents: it captures every tool call, prompt, response, and error from Claude Code, Cursor, or any MCP client, storing them for replay and analysis. In Python with zero runtime dependencies, you run CLI commands like `agent-strace setup` for Claude hooks, proxy MCP servers with `record --`, or wrap tools in decorators—then `replay`, `explain`, `stats`, or `cost` to debug sessions. It fills the gap where agents produce PRs without showing file reads, retries, or decision paths.

Why is it gaining traction?

Unlike LLM-only tracers, it grabs full agent loops—Bash commands, file edits, subagents—with timing, redaction for secrets, plain-English phase breakdowns spotting wasted retries, and token cost estimates flagging failed phases. Replay visualizes call chains like strace output; export to Datadog or Honeycomb via OTLP enables agent trace evaluation in production. The MCP proxy and Claude JSONL import make it instant for agent trace github users tired of opaque sessions.

Who should use this?

Devs iterating on Claude Code or Cursor agents debugging test failures or inefficient tool loops. Teams running MCP-based tools in Cursor/Windsurf needing capture without config hacks. Agent builders evaluating performance via visualization, cost breakdowns, or OTLP traces before scaling.

Verdict

Worth installing for any agent work—CLI shines, docs thorough despite 14 stars and 1.0% credibility score. Alpha maturity means test your flows, but it delivers core agent observability better than fragmented alternatives.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.