Prototype parses Claude's raw SSE stream into a forensic timeline of agent tool calls and diffs
Additional-Ladder-24 · reddit · 2026-10-28
- The author built a prototype that intercepts the SSE stream from claude.ai's web UI, parses the raw JSON, and generates a "Case Package" dashboard — no official API required.
- Test task: have the agent write a Python script that divides by zero, run it, fix it, and run again. The dashboard reconstructs exact causality: linking the bashtool call to the ZeroDivisionError, showing the strreplace diff used for the fix, and grouping failure/success evidence so an independent investigator can verify it.
- Motivation: current logging tools (CLI/web UI) show what an agent did but make it hard to see why it failed and which tool call caused a regression.
- The author asks who else is building a "Session Forensics" layer for agent debugging and what other events should be captured.
Related event: Dev Builds Agent Debugging Timeline by Intercepting Claude SSE Stream(3 posts)→
More from coding & agent
- Anne Griffin hosts 20-person SF salon on the product decisions behind AI agents — annetgriffin · 2026-10-28
- Coding agent caught rewriting its immutable requirements doc to pass failing tests — prasanna_says · 2026-10-28
- Adi Polak jokes about a 'CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS' env var — AdiPolak · 2026-10-28
- No coding background, 1 month with Claude + Unity: from AI models to a playable co-op game — PixelPawsInt · 2026-10-28
- DeepSeek-V4.1-Flash Q2 runs at 41 tok/s on one RTX 3090, 90% token match vs API — musicurgy · 2026-10-28
- Resend launches Inboxes, an inbox for agents, as developer marks 3-month sprint — philnash · 2026-10-28