Turn every AI chat into one searchable brain, then teach Claude to read it.

Tick these off as you go

• Claude Code installed (the CLI) — the engine that runs the parallel sub-agents
• Obsidian installed (free, from obsidian.md) — the app that browses your vault
• Python 3.10+ available on your shell — for the parser scripts
• ~500 MB free disk for a typical 1–2 year history
• About 60 minutes hands-on today, plus 30 minutes when your exports arrive

Scaffold the vault directory

• Open your terminal
• Run the scaffold command below to create the vault folders ↓
• Copy the scripts from the cloned repo into the vault's build directory

$ mkdir -p ~/ClaudeCodeBrain/{projects,topics,tools,entities,_build/scripts,_build/reports,_build/assignments}
$ cp ~/claude-brain-builder/scripts/*.py ~/ClaudeCodeBrain/_build/scripts/

Parse your terminal JSONL → markdown

• Run the export script — it walks ~/.claude/projects/, dedupes by working directory (worktrees collapse to their root), wraps tool calls in collapsible blocks, and writes one markdown file per session
• Wait for the "wrote N sessions" message — usually under a minute even for years of history

$ python3 ~/ClaudeCodeBrain/_build/scripts/export.py

1. Split sessions into 10 balanced slices

• Run the splitter — it does balanced bin-packing so all 10 agents finish at roughly the same time

$ python3 ~/ClaudeCodeBrain/_build/scripts/split_assignments.py

2. Dispatch 10 backlink agents inside Claude Code

• Open a Claude Code window (anywhere — your home directory works fine)
• Paste the orchestration prompt below ↓ — Claude will read the agent instructions and spawn all 10 sub-agents in parallel
• Wait for all 10 completion messages (~5–15 minutes wall-clock)

Spawn 10 parallel backlink agents using the Agent tool. Each agent gets a number 01-10. Brief each one with the instructions at ~/claude-brain-builder/prompts/backlink_terminal_agent.md and its assignment list at ~/ClaudeCodeBrain/_build/assignments/agent_NN.txt. Set run_in_background: true for all 10. Reports go to ~/ClaudeCodeBrain/_build/reports/agent_NN.json.

3. Consolidate the reports into hub pages

• Run the consolidate script — it reads all 10 JSON reports and writes one markdown page per topic, tool, and entity into topics/, tools/, entities/, plus a TOP-TOPICS.md leaderboard
• Run the rollup script — it walks the vault and rebuilds the project index + the top-level README from disk
• Open Obsidian → Open folder as vault → pick ~/ClaudeCodeBrain/ → confirm the README renders and the project index is populated

$ python3 ~/ClaudeCodeBrain/_build/scripts/consolidate.py
$ python3 ~/ClaudeCodeBrain/_build/scripts/rebuild_rollups.py

1. Drop the skill into Claude Code

• Create the skill directory and copy the skill file from the repo ↓
• Restart any running Claude Code sessions so the new skill loads

$ mkdir -p ~/.claude/skills/claude-brain
$ cp ~/claude-brain-builder/skill/claude-brain/SKILL.md ~/.claude/skills/claude-brain/

2. Drop the /slay sync command

• Copy the slay command file into your Claude Code commands directory ↓
• /slay will pull any new Claude Code sessions into the vault incrementally, without re-running the 10-agent backlink pass

$ mkdir -p ~/.claude/commands
$ cp ~/claude-brain-builder/commands/slay.md ~/.claude/commands/

3. Test it

• Open a fresh Claude Code window in any directory
• Type one of the trigger phrases (substitute a real project name) ↓
• Claude should pause, read the vault README and the project rollup, then summarize what you built before doing anything else
• If Claude just dives straight into code without reading first, double-check that ~/.claude/skills/claude-brain/SKILL.md exists

let's work on my-project

When the Claude.ai email arrives (~5 min)

• Download the ZIP from Anthropic's email and unzip it
• Run the importer below, pointing at the unzipped folder ↓
• Re-run the splitter for the web archive, then dispatch 10 more sub-agents (different prompt — see the repo)
• Re-run the consolidate script to merge the web archive into the same topic hubs as your terminal work

$ python3 ~/ClaudeCodeBrain/_build/scripts/import_claude_web.py /path/to/claude-export-dir
$ mkdir -p ~/ClaudeCodeBrain/_build/{web_assignments,web_reports}
$ python3 ~/ClaudeCodeBrain/_build/scripts/split_web_assignments.py

Spawn 10 parallel web backlink agents using the Agent tool. Each gets a number 01-10. Brief each with ~/claude-brain-builder/prompts/backlink_web_agent.md and its assignment list at ~/ClaudeCodeBrain/_build/web_assignments/agent_NN.txt. Reports go to ~/ClaudeCodeBrain/_build/web_reports/agent_NN.json. Set run_in_background: true for all 10.

$ python3 ~/ClaudeCodeBrain/_build/scripts/consolidate.py

When the ChatGPT email arrives (1–3 days)

• Download the ChatGPT ZIP from OpenAI's email and unzip it
• Run the ChatGPT importer below — it follows the displayed-thread chain and writes one markdown file per conversation, using the same frontmatter shape as the Claude.ai import
• Re-run the web splitter (it picks up the new files), dispatch 10 sub-agents with the same web backlink prompt, then re-run consolidate

$ python3 ~/ClaudeCodeBrain/_build/scripts/import_chatgpt.py /path/to/chatgpt-export-dir
$ python3 ~/ClaudeCodeBrain/_build/scripts/split_web_assignments.py

Quick reference