@toolbeltai/skills 0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+The MIT License
+
+Copyright (c) 2026 toolbeltai
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,64 @@
+# @toolbeltai/skills
+
+> Official Toolbelt skills for Claude Code and other MCP-capable agents —
+> SQL, vectors, graphs, geospatial, streaming, all as named slash commands.
+
+## Install (standalone)
+
+```bash
+npx @toolbeltai/skills install
+```
+
+Copies every skill to `~/.claude/skills/toolbelt/`. Restart Claude Code and
+you'll see them as slash commands: `/run-toolbelt`, `/geo-analyst`, …
+
+No account required; no network calls to Toolbelt. Skills work against any
+Toolbelt MCP server (cloud or self-hosted).
+
+## Install (via Toolbelt CLI)
+
+If you want the skills **and** a preconfigured MCP connection in one step:
+
+```bash
+npx @toolbeltai/cli
+```
+
+Provisions an anonymous Toolbelt account, registers the MCP server in your
+agent, and installs these skills — all at once.
+
+## Skills
+
+| Skill | Command | What it does |
+| --- | --- | --- |
+| [run-toolbelt](run-toolbelt/) | `/run-toolbelt` | Onboard, upload docs, connect data, ask questions |
+| [geo-analyst](geo-analyst/) | `/geo-analyst` | GPU-accelerated geospatial — queries and map rendering |
+| [knowledge-graph](knowledge-graph/) | `/knowledge-graph` | Auto-extract entities from docs, explore with Cypher |
+| [multi-agent-workspace](multi-agent-workspace/) | `/multi-agent-workspace` | Shareable MCP URL for multi-agent collaboration |
+| [sql-analyst](sql-analyst/) | `/sql-analyst` | Upload a CSV, ask plain English, get SQL + results |
+| [streaming-analyst](streaming-analyst/) | `/streaming-analyst` | Connect Kafka, aggregate, detect anomalies |
+| [vector-search](vector-search/) | `/vector-search` | Upload a document, retrieve semantically similar passages |
+| [data-blend](data-blend/) | `/data-blend` | Combine multiple tables with cross-table JOINs |
+
+## Works with
+
+Claude Code · OpenClaw · Cursor · Gemini CLI · Codex CLI · Windsurf · any MCP client
+
+## Commands
+
+```bash
+npx @toolbeltai/skills install       # install all skills
+npx @toolbeltai/skills uninstall     # remove them
+npx @toolbeltai/skills list          # list what would be installed
+npx @toolbeltai/skills path          # print install target
+```
+
+## Docs
+
+- Skill reference: <https://docs.toolbelt.ai>
+- Toolbelt CLI: <https://github.com/toolbeltai/toolbelt>
+- Releasing: [RELEASING.md](./RELEASING.md)
+- Contributing: [CONTRIBUTING.md](./CONTRIBUTING.md)
+
+## License
+
+MIT — see [LICENSE](./LICENSE).

package/assets/demos/fake-curl.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+sleep 1
+echo '{"mcpUrl": "https://mcp.toolbelt.ai/t/tb_1234567890/8bf91bad-02df-4137-97ac-69a006bac193/mcp"}'

package/assets/demos/geo-analyst-demo.sh

@@ -0,0 +1,85 @@
+#!/usr/bin/env bash
+# Simulates the geo-analyst skill running inside Claude Code
+
+G='\033[0;32m'   # green
+B='\033[0;34m'   # blue
+Y='\033[1;33m'   # yellow
+C='\033[0;36m'   # cyan
+D='\033[2;37m'   # dim
+W='\033[1;37m'   # bold white
+NC='\033[0m'
+
+printf "\n"
+printf "${D}> /geo-analyst${NC}\n"
+sleep 0.8
+
+printf "\n"
+printf "${C}● Phase 0  Verifying Toolbelt MCP connection...${NC}\n"
+sleep 0.6
+printf "${G}  ✓ Connected${NC}  |  namespace: demo-workspace\n"
+sleep 0.3
+
+printf "\n"
+printf "${C}● Phase 1  Resolving namespace...${NC}\n"
+sleep 0.4
+printf "${G}  ✓ Using${NC}  demo-workspace  (a3f2c1d9-4b2e-4a7c-9f1d-8e3b0c2d6a5f)\n"
+sleep 0.2
+
+printf "\n"
+printf "${C}● Phase 2  Uploading Tampa Bay sensor data...${NC}\n"
+sleep 0.4
+printf "  ${D}10 sensors  |  columns: id, name, lat, lon${NC}\n"
+sleep 0.8
+printf "  ${Y}[ingest]${NC}   running...\n"
+sleep 1.2
+printf "  ${G}[ingest]${NC}   completed ✓\n"
+sleep 0.3
+printf "${G}  ✓ Sensor table:${NC}  ${W}sensor_locations${NC}  (10 rows)\n"
+sleep 0.3
+
+printf "\n"
+printf "${C}● Phase 3  Running GPU geospatial queries...${NC}\n"
+sleep 0.3
+
+printf "\n"
+printf "  ${B}Query 1${NC}  Pairwise Distance  ${D}ST_DISTANCE()${NC}\n"
+sleep 1.0
+printf "  ┌────────────┬────────────┬────────────┐\n"
+printf "  │ sensor_a   │ sensor_b   │ distance_m │\n"
+printf "  ├────────────┼────────────┼────────────┤\n"
+printf "  │ Sensor A   │ Sensor B   │      2,847 │\n"
+printf "  │ Sensor A   │ Sensor J   │      4,213 │\n"
+printf "  │ Sensor B   │ Sensor J   │      5,091 │\n"
+printf "  └────────────┴────────────┴────────────┘\n"
+sleep 0.5
+
+printf "\n"
+printf "  ${B}Query 2${NC}  Point-in-Polygon  ${D}ST_CONTAINS()${NC}\n"
+sleep 0.9
+printf "  Zone: Downtown Tampa bounding polygon\n"
+printf "  ${G}  2 sensors inside zone:${NC}  Sensor A, Sensor B\n"
+sleep 0.5
+
+printf "\n"
+printf "  ${B}Query 3${NC}  Track Line  ${D}ST_MAKELINE()${NC}\n"
+sleep 0.9
+printf "  LINESTRING(-82.4572 27.9506, -82.4398 27.9659, ...)\n"
+printf "  ${G}  Track complete:${NC}  10 points\n"
+
+printf "\n"
+printf "${W}RESULT:${NC}\n"
+printf "  ${D}namespace_id:${NC}    a3f2c1d9-4b2e-4a7c-9f1d-8e3b0c2d6a5f\n"
+printf "  ${D}sensor_table:${NC}    sensor_locations  (10 rows)\n"
+printf "  ${D}phases_run:${NC}      [0, 1, 2, 3]\n"
+printf "\n"
+printf "  ${D}distance_query:${NC}\n"
+printf "    ${D}closest_pair:${NC}  Sensor A → Sensor B\n"
+printf "    ${D}min_distance:${NC}  2,847 m\n"
+printf "\n"
+printf "  ${D}point_in_polygon:${NC}\n"
+printf "    ${D}in_zone_count:${NC}   2\n"
+printf "    ${D}in_zone_sensors:${NC} [Sensor A, Sensor B]\n"
+printf "\n"
+printf "  ${D}track:${NC}\n"
+printf "    ${D}point_count:${NC}  10\n"
+printf "\n"

package/assets/demos/run-toolbelt-demo.sh

@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# Simulates the run-toolbelt skill running inside Claude Code
+
+G='\033[0;32m'   # green
+B='\033[0;34m'   # blue
+Y='\033[1;33m'   # yellow
+C='\033[0;36m'   # cyan
+D='\033[2;37m'   # dim
+W='\033[1;37m'   # bold white
+NC='\033[0m'
+
+printf "\n"
+printf "${D}> /run-toolbelt document_url=https://docs.toolbelt.ai/intro.pdf question=\"What is Toolbelt?\"${NC}\n"
+sleep 0.8
+
+printf "\n"
+printf "${C}● Phase 0  Verifying Toolbelt MCP connection...${NC}\n"
+sleep 0.6
+printf "${G}  ✓ Connected${NC}  |  namespace: demo-workspace  |  1 namespace found\n"
+sleep 0.3
+
+printf "\n"
+printf "${C}● Phase 1  Resolving namespace...${NC}\n"
+sleep 0.4
+printf "${G}  ✓ Using${NC}  demo-workspace  (a3f2c1d9-4b2e-4a7c-9f1d-8e3b0c2d6a5f)\n"
+sleep 0.2
+
+printf "\n"
+printf "${C}● Phase 2  Inspecting current state...${NC}\n"
+sleep 0.5
+printf "${G}  ✓ Context loaded${NC}  |  0 tables  |  0 vector collections\n"
+sleep 0.2
+
+printf "\n"
+printf "${C}● Phase 3  Uploading document...${NC}\n"
+sleep 0.4
+printf "  ${D}file:${NC} intro.pdf  ${D}source:${NC} docs.toolbelt.ai\n"
+sleep 0.8
+printf "  ${Y}[ingest]${NC}   running...\n"
+sleep 1.2
+printf "  ${G}[ingest]${NC}   completed ✓\n"
+sleep 0.4
+printf "  ${Y}[semantic]${NC} running...\n"
+sleep 1.6
+printf "  ${G}[semantic]${NC} completed ✓\n"
+sleep 0.3
+printf "${G}  ✓ Ingested${NC}  →  table: ${W}intro_pdf${NC}\n"
+sleep 0.3
+
+printf "\n"
+printf "${C}● Phase 5  Answering question...${NC}\n"
+sleep 0.4
+printf "  ${D}toolbelt_search  question=\"What is Toolbelt?\"${NC}\n"
+sleep 1.2
+printf "${G}  ✓ Done${NC}\n"
+
+printf "\n"
+printf "${W}RESULT:${NC}\n"
+printf "  ${D}namespace_id:${NC}   a3f2c1d9-4b2e-4a7c-9f1d-8e3b0c2d6a5f\n"
+printf "  ${D}phases_run:${NC}     [0, 1, 2, 3, 5]\n"
+printf "  ${D}document_table:${NC} intro_pdf\n"
+printf "  ${D}answer:${NC}         Toolbelt is a GPU-accelerated data workspace\n"
+printf "                  providing SQL, vector search, knowledge graphs,\n"
+printf "                  geospatial queries, and Kafka streaming via MCP.\n"
+printf "  ${D}sources:${NC}        [intro.pdf]\n"
+printf "\n"

package/bin/install.js

@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+/**
+ * toolbelt-skills — install Toolbelt's Claude Code skills into ~/.claude/skills/toolbelt/
+ *
+ * Usage:
+ *   npx @toolbeltai/skills install       # copy skills (default)
+ *   npx @toolbeltai/skills uninstall     # remove them
+ *   npx @toolbeltai/skills list          # print what would be installed
+ *   npx @toolbeltai/skills path          # print where they would go
+ *
+ * The @toolbeltai/cli package wraps this for its own install flow; this
+ * CLI exists so the skills package stands on its own — anyone can install
+ * without needing the Toolbelt CLI or hitting any Toolbelt-hosted service.
+ */
+import { cpSync, existsSync, mkdirSync, readdirSync, rmSync, statSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const PKG_ROOT = join(dirname(fileURLToPath(import.meta.url)), '..');
+const TARGET = join(homedir(), '.claude', 'skills', 'toolbelt');
+
+/** Any top-level directory containing a SKILL.md is a skill. */
+function listSkills() {
+  return readdirSync(PKG_ROOT, { withFileTypes: true })
+    .filter((e) => e.isDirectory() && !e.name.startsWith('.') && e.name !== 'node_modules' && e.name !== 'bin' && e.name !== 'assets')
+    .map((e) => e.name)
+    .filter((name) => existsSync(join(PKG_ROOT, name, 'SKILL.md')));
+}
+
+function cmdInstall() {
+  const skills = listSkills();
+  if (skills.length === 0) {
+    console.error('  ✗ No skills found in package. This is a packaging bug — please report.');
+    process.exit(1);
+  }
+
+  mkdirSync(TARGET, { recursive: true });
+  for (const name of skills) {
+    const src = join(PKG_ROOT, name);
+    const dst = join(TARGET, name);
+    rmSync(dst, { recursive: true, force: true });
+    cpSync(src, dst, { recursive: true });
+    console.log(`  \u2713 ${name}`);
+  }
+  // Optional assets (icons, tapes) — copy if present so popups render right.
+  const assetsSrc = join(PKG_ROOT, 'assets');
+  if (existsSync(assetsSrc) && statSync(assetsSrc).isDirectory()) {
+    const assetsDst = join(TARGET, 'assets');
+    rmSync(assetsDst, { recursive: true, force: true });
+    cpSync(assetsSrc, assetsDst, { recursive: true });
+  }
+
+  console.log('');
+  console.log(`  Installed ${skills.length} skills to ${TARGET}`);
+  console.log('  Restart Claude Code to pick them up.');
+}
+
+function cmdUninstall() {
+  if (!existsSync(TARGET)) {
+    console.log('  (nothing to remove)');
+    return;
+  }
+  rmSync(TARGET, { recursive: true, force: true });
+  console.log(`  Removed ${TARGET}`);
+}
+
+function cmdList() {
+  for (const name of listSkills()) console.log(`  ${name}`);
+}
+
+function cmdPath() {
+  console.log(TARGET);
+}
+
+function usage() {
+  console.log(`toolbelt-skills — install Toolbelt skills into ~/.claude/skills/toolbelt/
+
+Usage:
+  npx @toolbeltai/skills install       Install skills (default)
+  npx @toolbeltai/skills uninstall     Remove skills
+  npx @toolbeltai/skills list          List what would be installed
+  npx @toolbeltai/skills path          Print target install path
+`);
+}
+
+const cmd = process.argv[2] ?? 'install';
+switch (cmd) {
+  case 'install':
+    cmdInstall();
+    break;
+  case 'uninstall':
+    cmdUninstall();
+    break;
+  case 'list':
+    cmdList();
+    break;
+  case 'path':
+    cmdPath();
+    break;
+  case '-h':
+  case '--help':
+  case 'help':
+    usage();
+    break;
+  default:
+    console.error(`Unknown command: ${cmd}`);
+    usage();
+    process.exit(2);
+}

package/data-blend/SKILL.md

@@ -0,0 +1,283 @@
+---
+name: data-blend
+description: >
+  Join and correlate multiple datasets in a single Toolbelt namespace without
+  writing infrastructure code. Toolbelt is a multi-modal data platform combining
+  SQL analytics, vector search, and real-time streaming. Uploads two or more CSV
+  tables, then runs cross-table JOIN queries to surface relationships between
+  datasets. Use when an AI agent needs to combine data from different sources —
+  orders with customers, sensors with metadata, events with dimensions — and
+  answer questions that span multiple tables.
+license: MIT
+compatibility: >
+  Requires a Toolbelt account (provision free at https://toolbelt.ai) and an
+  MCP-compatible AI agent (Claude Code, Claude Desktop, or any client that
+  supports MCP server connections). MCP connection must be pre-established
+  before invocation.
+metadata:
+  author: toolbeltai
+  version: "1.0"
+  openclaw:
+    emoji: "🔀"
+    homepage: "https://toolbelt.ai/docs/sql"
+    skillKey: "data-blend"
+---
+
+Upload multiple tables and run cross-table JOIN queries using Toolbelt MCP tools.
+Work through each phase in order without prompting for user input. On
+unrecoverable error, emit a structured failure and halt.
+
+## When Not To Use
+
+- For a single table — use `sql-analyst` instead.
+- For unstructured text or documents — use `knowledge-graph` instead.
+- For streaming/real-time data — use `streaming-analyst` instead.
+- For spatial data with lat/lon — use `geo-analyst` instead.
+
+## Invocation Parameters
+
+Extract these from the args string or conversation context before starting:
+
+| Parameter | Required | Description |
+|---|---|---|
+| `namespace_id` | No | UUID of target namespace. Auto-select if omitted and only one exists; fail if ambiguous. |
+| `table_a_content` | No | Raw CSV for the first table. Uses default `orders` sample if omitted. |
+| `table_a_name` | No | Asset name for the first table. Defaults to `orders`. |
+| `table_b_content` | No | Raw CSV for the second table. Uses default `customers` sample if omitted. |
+| `table_b_name` | No | Asset name for the second table. Defaults to `customers`. |
+| `join_query` | No | Custom SQL JOIN to execute in Phase 5. Uses default query if omitted. |
+| `skip_upload` | No | Set to `true` to skip Phases 2–4 and query tables already in the namespace. |
+
+---
+
+## Default Sample Data
+
+If no `table_a_content` is provided, use this orders dataset verbatim:
+
+```
+order_id,customer_id,product,category,quantity,amount,order_date
+1001,C001,Widget Pro,Hardware,12,599.88,2024-01-05
+1002,C003,Gadget Basic,Software,5,149.95,2024-01-08
+1003,C002,Widget Pro,Hardware,8,399.92,2024-01-12
+1004,C004,Service Plan,Services,3,597.00,2024-01-15
+1005,C001,Gadget Basic,Software,20,599.80,2024-01-19
+1006,C005,Widget Pro,Hardware,6,299.94,2024-01-22
+1007,C003,Service Plan,Services,2,398.00,2024-02-03
+1008,C002,Gadget Plus,Software,15,1199.85,2024-02-07
+1009,C001,Widget Pro,Hardware,10,499.90,2024-02-11
+1010,C004,Gadget Basic,Software,8,239.92,2024-02-14
+1011,C005,Gadget Plus,Software,4,319.96,2024-02-18
+1012,C002,Service Plan,Services,1,199.00,2024-02-21
+1013,C001,Service Plan,Services,5,995.00,2024-03-02
+1014,C005,Gadget Plus,Software,9,719.91,2024-03-06
+1015,C003,Widget Pro,Hardware,7,349.93,2024-03-10
+1016,C002,Gadget Basic,Software,11,329.89,2024-03-14
+1017,C004,Gadget Plus,Software,6,479.94,2024-03-18
+1018,C005,Service Plan,Services,4,796.00,2024-03-22
+1019,C003,Widget Pro,Hardware,3,149.97,2024-03-25
+1020,C002,Widget Pro,Hardware,14,699.86,2024-03-28
+```
+
+If no `table_b_content` is provided, use this customers dataset verbatim:
+
+```
+customer_id,name,region,segment,account_manager
+C001,Meridian Corp,Northeast,Enterprise,Alice Chen
+C002,Delta Systems,Midwest,Enterprise,Carol Singh
+C003,Apex Solutions,Southeast,Mid-Market,Bob Martinez
+C004,Crest Industries,West,Mid-Market,David Park
+C005,Solaris Group,West,SMB,Emma Lopez
+```
+
+Default `join_query`:
+```sql
+SELECT
+  c.segment,
+  c.region,
+  COUNT(o.order_id) AS order_count,
+  ROUND(SUM(o.amount), 2) AS total_amount,
+  ROUND(AVG(o.amount), 2) AS avg_order_value
+FROM <orders_table> o
+JOIN <customers_table> c ON o.customer_id = c.customer_id
+GROUP BY c.segment, c.region
+ORDER BY total_amount DESC
+```
+
+---
+
+## Phase 0: Verify Connection
+
+Call `get_semantic_names` (no arguments) immediately.
+
+- **If it succeeds:** proceed to Phase 1 using the returned namespaces.
+- **If it fails:** emit structured failure and halt.
+
+```
+FAILURE: Toolbelt MCP connection is not established.
+The MCP server must be connected before invoking this skill.
+See: https://toolbelt.ai/docs/mcp for setup instructions.
+```
+
+---
+
+## Phase 1: Resolve Namespace
+
+Use the namespaces returned from Phase 0.
+
+Resolution order:
+1. If `namespace_id` was provided as a parameter, use it directly.
+2. If only one namespace exists, use it.
+3. If multiple exist and no `namespace_id` was specified, emit structured failure and halt.
+
+```
+FAILURE: Multiple namespaces found and none specified.
+Available: [<list namespace display names and IDs>]
+Re-invoke with namespace_id=<uuid>.
+```
+
+Store the resolved `namespace_id` — pass it to every subsequent tool call.
+
+---
+
+## Phase 2: Inspect Existing Tables
+
+Skip this phase if `skip_upload` is `true` — go directly to Phase 5.
+
+Call `toolbelt_context` with `{ "namespace_id": "<namespace_id>" }`.
+
+Check whether tables matching `table_a_name` and `table_b_name` already exist:
+- If both exist: skip Phases 3–4 and proceed to Phase 5 using their existing table names.
+- If one or both are missing: upload the missing ones in Phase 3.
+
+Store all resolved table names for use in Phase 5.
+
+---
+
+## Phase 3: Upload Missing Tables
+
+For each table that does not already exist, call `toolbelt_save`:
+
+```json
+{
+  "asset_type": "document",
+  "namespace_id": "<namespace_id>",
+  "name": "<table_a_name or table_b_name>",
+  "file_name": "<name>.csv",
+  "content": "<csv_content>",
+  "content_encoding": "text",
+  "data_format": "csv"
+}
+```
+
+Record each returned `asset_id`.
+
+---
+
+## Phase 4: Poll for Ingestion
+
+Call `toolbelt_jobs` with `{ "namespace_id": "<namespace_id>" }` every 10 seconds.
+
+Wait for the `ingest` job for each uploaded asset to reach `completed`. Typical duration: 15–60 seconds per table. Maximum wait: 3 minutes.
+
+If any job reaches `failed` or the timeout elapses, emit structured failure and halt:
+```
+FAILURE: Table ingestion did not complete.
+Asset: <table name>
+Job status: <last observed status>
+```
+
+After all jobs complete, call `toolbelt_context` to retrieve the final SQL table names for all assets. Store as `table_a` and `table_b`.
+
+---
+
+## Phase 5: Run Blend Queries
+
+Substitute `<orders_table>` and `<customers_table>` (or equivalent names) in all queries with the resolved table names from Phase 4 (or Phase 2 if skip_upload).
+
+### Query 1 — Join summary (default or custom)
+
+If `join_query` was provided, execute it directly. Otherwise execute the default query, substituting the resolved table names:
+
+```json
+{
+  "namespace_id": "<namespace_id>",
+  "query": "<join_query with table names substituted>"
+}
+```
+
+Use `toolbelt_sql` for this call. Record:
+- `join_rows`: number of rows returned
+- `join_results`: up to 10 rows
+
+### Query 2 — Row counts
+
+Confirm both tables are populated:
+
+```sql
+SELECT '<table_a>' AS table_name, COUNT(*) AS row_count FROM <table_a>
+UNION ALL
+SELECT '<table_b>' AS table_name, COUNT(*) AS row_count FROM <table_b>
+```
+
+Record `row_counts` for each table.
+
+### Query 3 — Unmatched rows check
+
+Identify rows in table A with no match in table B (join integrity check):
+
+```sql
+SELECT COUNT(*) AS unmatched_count
+FROM <table_a> a
+LEFT JOIN <table_b> b ON a.<join_key> = b.<join_key>
+WHERE b.<join_key> IS NULL
+```
+
+Infer `<join_key>` from the shared column names visible in the schema context. If no shared key can be inferred, skip this query and note it in the RESULT.
+
+Record `unmatched_count`.
+
+---
+
+## Phase 6: Structured Output
+
+After all phases complete, emit a single structured result:
+
+```
+RESULT:
+  namespace_id: <uuid>
+  phases_run: [0, 1, 2, 3, 4, 5]
+
+  tables:
+    table_a: <sql table name>
+    table_b: <sql table name>
+    row_counts:
+      <table_a>: <count>
+      <table_b>: <count>
+    unmatched_rows: <count or "skipped — no shared key inferred">
+
+  blend_query:
+    sql: |
+      <query executed>
+    row_count: <join_rows>
+    results:
+      - <row 1>
+      - <row 2>
+      ... (up to 10 rows)
+```
+
+If any Phase 5 query fails, include `query_error: "<error>"` under that query's
+section and continue. Only halt on Phase 0–3 failures.
+
+---
+
+## Tool Reference
+
+| Phase | Tool(s) |
+|---|---|
+| 0. Verify connection | `get_semantic_names` |
+| 1. Resolve namespace | (from Phase 0 result) |
+| 2. Inspect existing tables | `toolbelt_context` |
+| 3. Upload missing tables | `toolbelt_save` |
+| 4. Poll for ingestion | `toolbelt_jobs`, `toolbelt_context` |
+| 5. Run blend queries | `toolbelt_sql`, `toolbelt_execute` |
+| 6. Emit result | (structured output) |

package/geo-analyst/README.md

@@ -0,0 +1,7 @@
+# geo-analyst
+
+![geo-analyst demo](../assets/geo-analyst-demo.gif)
+
+GPU-accelerated geospatial analytics agent powered by Toolbelt MCP. Uploads lat/lon sensor data, runs geospatial SQL queries (distance, point-in-polygon, track creation), and emits structured results.
+
+Invoke via `/geo-analyst` in Claude Code, or via the `Skill` tool in any MCP-capable agent.