What Happened
A developer released Argus, an open-source VSCode extension that parses Claude Code's local session logs and presents them through an eight-tab analysis dashboard. The extension scans your ~/.claude/projects/ directory, discovers sessions automatically, and gives you detailed breakdowns of what happened during each coding session.
The eight tabs cover a lot of ground. The Overview tab shows session statistics and cost summaries. The Analysis tab runs six detection rules to find duplicate reads, unused operations, retry loops, failed tools, context pressure, and compaction events. The Cost tab provides step-by-step token breakdowns with cache hit ratios and model attribution. The Performance tab calculates efficiency scores and identifies wasted spend.
The more interesting tabs are Flow and Context. Flow uses D3.js to render interactive dependency graphs showing how file operations relate to each other across steps. Context tracks token usage and cache performance so you can see exactly how your context window was utilized throughout a session.
The tech stack is TypeScript with React 19.2 for the webview, Vite for bundling, Chart.js and D3.js for visualizations. It requires Node.js 18+ and can be installed via VSIX package or built from source.
Why It Matters
Claude Code's biggest blind spot right now is observability. You know roughly what it costs per session from your Anthropic dashboard, but you have no granular visibility into where tokens went, which operations were wasteful, or where the agent got stuck in retry loops.
Argus fills that gap. The retry loop detection alone is valuable. Anyone who has watched Claude Code attempt the same failing operation three times in a row knows the frustration. Being able to see those patterns after the fact lets you adjust your workflows and prompts to avoid them.
The cost tracking is equally practical. If you are on a team where multiple developers use Claude Code daily, understanding per-session cost breakdowns and cache hit ratios helps you budget and optimize usage.
Our Take
This is the kind of tooling that should exist but does not get built by the platform vendors. Anthropic has little incentive to show you exactly where you are wasting tokens. The community had to build it.
The six-rule analysis engine is the standout feature. Detecting duplicate file reads, unused operations, and context pressure gives you actionable data to improve how you use Claude Code. Most people treat AI coding sessions as black boxes. Argus cracks them open.
The one limitation is that this only works with Claude Code's local JSONL logs. If Anthropic changes the log format, Argus breaks. That is the inherent risk of building on undocumented internals. But for now, if you use Claude Code regularly and want to understand where your money goes, Argus is worth installing. It is free, open-source, and the eight-tab dashboard is genuinely more informative than anything Anthropic provides natively.