---
name: gentic-architect
description: "Give your AI agent a data architect. Scan a Google Drive folder, get a reviewable import plan, and load clean tables and searchable knowledge into your Gentic workspace on approval — all through the Model Context Protocol."
license: MIT
metadata:
  author: gentic
  version: "1.0.0"
---

# Gentic Data Architect

Point your agent at a Google Drive folder. The Data Architect reads every doc, PDF, and spreadsheet, proposes an import plan — what becomes a queryable table, what becomes searchable knowledge, and what needs a human — then, on your approval, converts and loads it all into your Gentic workspace. Scan, plan, apply — all through the Model Context Protocol.

## When to apply

- User wants to import a Google Drive folder of files into AI-ready data.
- User has spreadsheets, PDFs, docs, or CSVs that need to become queryable tables or searchable knowledge.
- User wants to review what will be imported — and why — before anything runs.
- User wants to re-run a folder after updating files, refreshing the data in place.
- User asks what happened to a specific file in a previous import (imported, skipped, or flagged).

## Tools

| Tool | Description | Cost |
|------|-------------|------|
| `architect_apply` | Apply an approved Data Architect plan: approve/reject/skip each item, resolving any conflicts (table_name_override). Approved items only; the review gate is structural and validated atomically — any unresolved conflict rejects the whole submission and nothing changes. Import of approved items runs server-side. | Free |
| `architect_get_plan` | Read the review plan for a Data Architect scan: per-file destination, proposed spec, conflicts to resolve, and current state. Approve/reject items with architect_apply. | Free |
| `architect_scan_folder` | Onboard a whole Google Drive folder into GenticDB: classify every file (brain doc vs data table vs hybrid vs skip vs needs-human) into a REVIEW PLAN. ASYNC — returns a scan_id immediately with status 'scanning'; poll architect_get_plan(scan_id) until status is 'planned' (a folder usually takes ~1-2 min). Nothing is written until you approve the plan with architect_apply. Use this for 'onboard this Drive folder' / 'sort these files into brain vs tables' — a single 'import this CSV' or 'add this doc to the brain' stays on the data/brain tools. Billed 5¢ per file classified, only once the scan completes. | 5¢ / result (min 5¢) |

## Workflow

### 1. Scan the folder first

Call `architect_scan_folder` with a Google Drive folder link. It enumerates the files it can read, inspects each one, and produces an import plan: for every file a proposed destination (queryable table, searchable knowledge, table + search, or needs-review), plus the structure it detected — header row, which tab, which columns should get semantic search. Ambiguous files are flagged with a specific reason rather than guessed at. Nothing is imported at this stage. Surface the returned plan to the user before applying.

### 2. Review the plan before applying

Use `architect_get_plan` to fetch the current plan for a scanned folder (e.g. to re-display it, or after the user steps away). Walk the user through it file by file — especially anything flagged needs-review, which won't import until the user decides. The plan is the product: it's the gate the user approves against.

### 3. Apply only approved files

`architect_apply` runs the import for the approved files and converts each into its planned destination in the user's Gentic workspace: tables you can query in plain English, documents your AI can search and cite. It returns a receipt — what landed, what was skipped, and why. Imports are enforced server-side against the approved plan, so an agent cannot import a file the user didn't approve.

### 4. Re-run safely to refresh

If files change, re-scan and re-apply the folder. Changed files refresh their tables in place rather than creating duplicates, so re-running is safe and idempotent.

### 5. Present results clearly

Don't dump raw JSON. Show the plan as a readable per-file list (filename → destination → one-line detail), call out flagged files with their reason, and after apply, summarize the receipt: how many files landed as tables, how many as knowledge, and what was skipped.

## Notes

- Your Drive is read-only to the Architect — it reads files to plan and import; it never modifies your Drive. Converted copies live in your Gentic workspace with a full audit trail.
- Nothing imports without approval: `architect_apply` only acts on files approved in the plan, enforced server-side.
- Two separate meters: scanning a folder (architect_scan_folder) is billed per file inspected, whether or not it ends up imported; applying a plan (architect_apply) is billed per file actually imported — files you skip or that fail cost nothing on apply. Exact per-tool pricing is pulled live from the Gentic MCP manifest below.
- Ambiguous files are never guessed at — they're flagged in the plan with a specific reason and a suggested fix, and left for a human to decide.
- Supported file types include Google Docs, Google Sheets, Excel workbooks, PDFs, CSVs, and plain text files.

## Tool details

- `architect_apply` — Apply an approved Data Architect plan: approve/reject/skip each item, resolving any conflicts (table_name_override). Approved items only; the review gate is structural and validated atomically — any unresolved conflict rejects the whole submission and nothing changes. Import of approved items runs server-side.
  - `scan_id` (string, required)
  - `approvals` (array of object, required) — Per-item decisions. Duplicate-name conflicts MUST be resolved here (table_name_override or skip).
- `architect_get_plan` — Read the review plan for a Data Architect scan: per-file destination, proposed spec, conflicts to resolve, and current state. Approve/reject items with architect_apply.
  - `scan_id` (string, required) — The scan_id returned by architect_scan_folder.
- `architect_scan_folder` — Onboard a whole Google Drive folder into GenticDB: classify every file (brain doc vs data table vs hybrid vs skip vs needs-human) into a REVIEW PLAN. ASYNC — returns a scan_id immediately with status 'scanning'; poll architect_get_plan(scan_id) until status is 'planned' (a folder usually takes ~1-2 min). Nothing is written until you approve the plan with architect_apply. Use this for 'onboard this Drive folder' / 'sort these files into brain vs tables' — a single 'import this CSV' or 'add this doc to the brain' stays on the data/brain tools. Billed 5¢ per file classified, only once the scan completes.
  - `folder_url` (string) — Public 'anyone with the link' Google Drive folder URL.
  - `file_urls` (array of string) — Fallback when folder enumeration is blocked: explicit public file links.
  - `max_files` (integer) — Hard cap (default 100).

---

_This SKILL.md is generated from the live Gentic MCP manifest. Tool names, descriptions, and pricing are always current. Connect Gentic Data Architect at https://gentic.co/architect._