Notes from the Field: Building a Second Brain from Claude Code Conversations

This is an abstract of a Chinese-language Synapse article. Every day’s Claude Code conversations contain decisions, problem solutions, learnings, and project notes that vanish when the session closes. The full Chinese article documents a Software 2.0 system that extracts conversation .jsonl files nightly, filters system noise, classifies user intent into five high-value knowledge categories (decisions / concepts / problems / learnings / projects), and writes the result into an Obsidian-based second brain. The system runs at 23:00 daily and produces ~30–40 high-value knowledge items per ~100 messages.

Key Takeaways

• Full conversation extraction beats daily summaries: pulling from .jsonl files achieves ~100% coverage vs. ~0.2% for hand-written daily-record approaches.
• Noise filtering is mandatory: strip system-reminder, tool results, and meta-prompts; preserve only the user’s original text and real intent. Otherwise the AI signal-to-noise ratio drops below useful threshold.
• Five-category intent classifier: decision (approvals, plans), concept (new methods, frameworks), problem (errors and fixes), learning (new understanding), project (named projects). Each category has its own keyword anchors to bootstrap classification.
• Software 2.0 vs 1.0: traditional knowledge management requires human tagging and hand-built links. Software 2.0 lets AI learn implicit rules from usage patterns, evolving relevance over time.

Why This Matters

Most teams using Claude Code (or any chat-driven AI tool) treat conversations as ephemeral. That’s a massive ongoing knowledge leak. The Synapse system flips it: conversations become the primary training input for an organization’s evolving knowledge graph. The Obsidian-as-second-brain pattern means the same knowledge surfaces in future sessions via Claude’s read-access — every conversation makes the next one smarter. The Chinese article includes the toolchain (Claude Code, Python parser, Obsidian, n8n, Lysander as orchestrator) and concrete extraction stats. If your team generates more conversation than documentation, this is the architecture that captures it.

---

This is an abstract. Read the full article in Chinese → 如何用 AI 打造第二大脑