add global CLAUDE.md: sql style, query tools, nvim awareness

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

_lib.sh

@@ -52,6 +52,9 @@ deploy_configs() {
     migrate_and_link "$config_dir/$config" ~/"$config"
   done
 
+  mkdir -p ~/.claude
+  create_symlink "$config_dir/.claude/CLAUDE.md" ~/.claude/CLAUDE.md
+
   echo "Sourcing .bashrc..."
   source ~/.bashrc
 }

dotfiles/.claude/CLAUDE.md

@@ -0,0 +1,50 @@
+# Global Claude Preferences
+
+## SQL Style
+
+Always write SQL in this style:
+
+- **Keywords UPPERCASE**: SELECT, FROM, WHERE, WITH, GROUP BY, LEFT JOIN, ON, CASE, WHEN, COALESCE, SUM, MAX, etc.
+- **Leading commas**, one column per line, indented under SELECT/GROUP BY
+- **FROM / WHERE / GROUP BY** on their own line, content indented below
+- **LEFT JOIN ... ON** with condition indented on next line
+- **WHERE TRUE** as base condition
+- **CTEs** with `--` comment header; subsequent CTEs start with `,cte_name AS (`
+- `::TYPE` casting over `CAST()`
+- `~` for regex, not LIKE
+- No trailing commas, no unnecessary parentheses
+- Descriptive snake_case aliases
+
+Canonical example: `inventory/on_hand/onhand_with_last_mfg.pg.sql` in the SQL repo.
+
+## Shell Query Tools
+
+PostgreSQL (connection string in `$PG`, defined in `.bashrc_local`):
+- `xnspa` — fzf-select open nvim buffer, run with `$PG`, output to VisiData
+- `xnsp` — same, output to pspg
+- `xns` — same, raw output
+
+DuckDB (`/opt/sync/osm_stack.duckdb` snapshot, CLI at `~/.local/bin/duckdb`):
+- `xnd` — fzf-select open nvim buffer, run against DuckDB snapshot, output to VisiData
+- Direct: `duckdb /opt/sync/osm_stack.duckdb < file.du.sql`
+
+SQL Server (connection strings `$MS`, `$MSC`, `$MSW` in `.bashrc_local`):
+- `xmspa` — fzf-select open nvim buffer, run with `$MSC`, output to VisiData
+
+File extension conventions:
+- `.pg.sql` — PostgreSQL
+- `.ms.sql` — SQL Server
+- `.db2.sql` — IBM DB2
+- `.du.sql` — DuckDB
+
+## Nvim Buffer Awareness
+
+At the start of every conversation, check for open nvim buffers in the current directory and read them proactively:
+
+```bash
+lsof +D ~/.local/state/nvim/swap/ 2>/dev/null | grep -o "/swap/.*" | sed 's|^/swap/||' | tr "%" "/" | sed -E 's/\.sw[op]$//' | grep "$(pwd | sed 's|^//|/|')"
+```
+
+## Setup Environment
+
+Dotfiles are managed via symlinks from `~/setup_env/dotfiles/`. Editing `~/.<file>` directly edits the repo. Always edit dotfiles at their source (`~/setup_env/dotfiles/`) rather than the symlink target. `_lib.sh` handles deployment; `.bashrc_local` and `.bashrc_paths` are gitignored and machine-specific.