synclisted

# Context Sync — Cross-Device + Cross-Tool Two problems, one canonical source of truth: 1. **Cross-device** — Claude Code identifies projects by a hash of the absolute path, so context is lost between machines (`/home/alice/myapp` → `-home-alice-myapp` ≠ `/Users/john/myapp` → `-Users-john-myapp`). Solved by `export` / `import`. 2. **Cross-tool** — You use Claude Code, Cursor, Codex CLI, Roo, Cline, Aider, or Windsurf — each has its own instruction/memory file (`CLAUDE.md`, `.cursorrules`, `AGENTS.md`, ...). You re-explain your project to every one. Solved by `tools push` — one canonical memory file → adapter files for every detected tool. **The shared truth:** `.claude-context/MEMORY.md` (+ `memory/*.md`), committed to git. Every tool reads from a locally-generated adapter file pointing at that truth. No SaaS. No proprietary memory broker. No vendor lock-in. --- ## Commands ### `/sync export` Export current project context to `.claude-context/` and commit. Run: ```bash python scripts/sync-export.py ``` What it does: 1. Calculates the Claude path hash for the current directory 2. Finds `~/.claude/projects/<hash>/` — reads all `.md` context files 3. Copies them to `.claude-context/` in the project 4. Creates `.claude-context/sync-manifest.json` with device metadata 5. Runs `git add .claude-context/ && git commit -m "context-sync: export <timestamp>"` 6. Prints: push reminder Then push: ```bash git push ``` **What gets exported:** - `MEMORY.md` — your project memory