db-mcp 1.1.1 → 3.0.0

package/test-server/reset-database.ps1

@@ -1,373 +0,0 @@
-#!/usr/bin/env pwsh
-<#
-.SYNOPSIS
-    Resets the db-mcp test database by deleting and re-creating from seed data.
-
-.DESCRIPTION
-    This script performs a full reset of the db-mcp test database:
-    1. Deletes the existing test database file (if exists)
-    2. Creates a fresh database from test-database.sql
-    3. Verifies table counts and data integrity
-    4. Displays a summary of seeded data
-
-.PARAMETER DatabasePath
-    Path to the database file. Default: test.db in the script directory.
-
-.PARAMETER SkipVerify
-    Skip the verification step after reset.
-
-.PARAMETER Verbose
-    Show detailed output for each step.
-
-.EXAMPLE
-    .\reset-database.ps1
-
-.EXAMPLE
-    .\reset-database.ps1 -DatabasePath "mytest.db" -Verbose
-
-.EXAMPLE
-    .\reset-database.ps1 -SkipVerify
-#>
-
-param(
-    [string]$DatabasePath = "",
-    [switch]$SkipVerify,
-    [switch]$Verbose
-)
-
-$ErrorActionPreference = "Stop"
-$ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path
-$SqlFile = Join-Path $ScriptDir "test-database.sql"
-
-# Default database path
-if (-not $DatabasePath) {
-    $DatabasePath = Join-Path $ScriptDir "test.db"
-}
-
-# Colors for output
-function Write-Step { param($Step, $Total, $Message) Write-Host "`n[$Step/$Total] " -ForegroundColor Cyan -NoNewline; Write-Host $Message -ForegroundColor White }
-function Write-Success { param($Message) Write-Host "  [OK] " -ForegroundColor Green -NoNewline; Write-Host $Message }
-function Write-Info { param($Message) Write-Host "  --> " -ForegroundColor DarkGray -NoNewline; Write-Host $Message -ForegroundColor DarkGray }
-function Write-Warn { param($Message) Write-Host "  [!] " -ForegroundColor Yellow -NoNewline; Write-Host $Message -ForegroundColor Yellow }
-function Write-Err { param($Message) Write-Host "  [X] " -ForegroundColor Red -NoNewline; Write-Host $Message -ForegroundColor Red }
-
-Write-Host "`n========================================================" -ForegroundColor Magenta
-Write-Host "            DB-MCP Test Database Reset                   " -ForegroundColor Magenta
-Write-Host "========================================================" -ForegroundColor Magenta
-
-# Verify prerequisites
-if (-not (Test-Path $SqlFile)) {
-    Write-Err "test-database.sql not found at: $SqlFile"
-    exit 1
-}
-Write-Host "`nScript directory: " -NoNewline; Write-Host $ScriptDir -ForegroundColor Gray
-Write-Host "Database path: " -NoNewline; Write-Host $DatabasePath -ForegroundColor Gray
-Write-Host "SQL seed file: " -NoNewline; Write-Host $SqlFile -ForegroundColor Green
-
-# Check if sqlite3 is available
-$sqlite3Path = $null
-try {
-    $sqlite3Path = Get-Command sqlite3 -ErrorAction Stop | Select-Object -ExpandProperty Source
-} catch {
-    # Try common locations on Windows
-    $commonPaths = @(
-        "C:\sqlite\sqlite3.exe",
-        "C:\Program Files\sqlite\sqlite3.exe",
-        "$env:LOCALAPPDATA\Programs\sqlite\sqlite3.exe"
-    )
-    foreach ($path in $commonPaths) {
-        if (Test-Path $path) {
-            $sqlite3Path = $path
-            break
-        }
-    }
-}
-
-if (-not $sqlite3Path) {
-    Write-Warn "sqlite3 not found in PATH. Will use Node.js better-sqlite3 via db-mcp CLI."
-    $useCli = $true
-} else {
-    Write-Host "SQLite3 path: " -NoNewline; Write-Host $sqlite3Path -ForegroundColor Green
-    $useCli = $false
-}
-
-$totalSteps = if ($SkipVerify) { 2 } else { 3 }
-
-# ============================================================================
-# Step 1: Clean up test artifacts and delete existing database
-# ============================================================================
-Write-Step "1" $totalSteps "Cleaning up test artifacts and deleting database..."
-
-# Remove backup .db files left by sqlite_backup tests
-$backupFiles = Get-ChildItem -Path $ScriptDir -Filter "*.db" -ErrorAction SilentlyContinue |
-    Where-Object { $_.Name -ne "test.db" }
-if ($backupFiles) {
-    foreach ($bf in $backupFiles) {
-        Remove-Item $bf.FullName -Force -ErrorAction SilentlyContinue
-        if ($Verbose) { Write-Info "Deleted backup: $($bf.Name)" }
-    }
-    Write-Success "Cleaned up $($backupFiles.Count) backup file(s)"
-}
-
-# Warn about any node processes that hold the database file open (e.g. MCP servers)
-$dbFullPath = (Resolve-Path $DatabasePath -ErrorAction SilentlyContinue)?.Path
-if (-not $dbFullPath) { $dbFullPath = $DatabasePath }
-$dbNameEscaped = [regex]::Escape(($dbFullPath -replace '\\', '/'))
-$dbNameEscapedBackslash = [regex]::Escape($dbFullPath)
-
-$nodeProcesses = Get-CimInstance Win32_Process -Filter "Name = 'node.exe'" -ErrorAction SilentlyContinue |
-    Where-Object { $_.CommandLine -match $dbNameEscaped -or $_.CommandLine -match $dbNameEscapedBackslash }
-
-if ($nodeProcesses) {
-    foreach ($proc in $nodeProcesses) {
-        Write-Warn "Node process PID $($proc.ProcessId) is using this database (likely an MCP server)"
-    }
-    Write-Info "The database file will be replaced in-place — existing connections will pick up the new data."
-}
-
-$filesToDelete = @(
-    $DatabasePath,
-    "$DatabasePath-shm",
-    "$DatabasePath-wal",
-    "$DatabasePath-journal"
-)
-
-$deletedCount = 0
-$lockFailed = $false
-foreach ($file in $filesToDelete) {
-    if (Test-Path $file) {
-        try {
-            Remove-Item $file -Force -ErrorAction Stop
-            $deletedCount++
-            if ($Verbose) {
-                Write-Info "Deleted: $file"
-            }
-        } catch {
-            if ($file -eq $DatabasePath) {
-                # Main DB file is locked — overwrite will happen at seed step
-                Write-Warn "Database file is locked; will overwrite in-place at seed step"
-                $lockFailed = $true
-            } else {
-                Write-Info "Could not delete $file (may be in use): $_"
-            }
-        }
-    }
-}
-
-if ($deletedCount -gt 0) {
-    Write-Success "Deleted $deletedCount file(s)"
-} elseif (-not $lockFailed) {
-    Write-Info "No existing database files found"
-}
-
-# ============================================================================
-# Step 2: Create fresh database from seed file
-# ============================================================================
-Write-Step "2" $totalSteps "Creating database from seed file..."
-
-if (-not $useCli) {
-    # Use sqlite3 command line with input redirection
-    # The .read command with -cmd doesn't work reliably, so we use Get-Content | sqlite3
-    try {
-        Get-Content $SqlFile -Raw | & $sqlite3Path $DatabasePath
-        if ($LASTEXITCODE -ne 0) {
-            throw "sqlite3 exited with code $LASTEXITCODE"
-        }
-        Write-Success "Database created using sqlite3"
-    } catch {
-        Write-Err "Failed to create database: $_"
-        exit 1
-    }
-} else {
-    # Use Node.js with better-sqlite3 via inline script
-    $dbMcpRoot = Split-Path -Parent $ScriptDir
-    $nodeScript = @"
-import Database from 'better-sqlite3';
-import { readFileSync } from 'fs';
-
-const dbPath = process.argv[2];
-const sqlPath = process.argv[3];
-
-try {
-    const db = new Database(dbPath);
-    const sql = readFileSync(sqlPath, 'utf-8');
-    db.exec(sql);
-    db.close();
-    console.log('SUCCESS');
-} catch (err) {
-    console.error('ERROR:', err.message);
-    process.exit(1);
-}
-"@
-
-    # Write temp script to project directory so node can resolve dependencies
-    $tempScript = Join-Path $dbMcpRoot ".create-test-db.js"
-    $nodeScript | Out-File -FilePath $tempScript -Encoding utf8 -NoNewline
-
-    Push-Location $dbMcpRoot
-    try {
-        $result = node $tempScript $DatabasePath $SqlFile 2>&1
-        if ($LASTEXITCODE -ne 0 -or $result -notmatch "SUCCESS") {
-            Write-Err "Failed to create database: $result"
-            exit 1
-        }
-        Write-Success "Database created using Node.js (better-sqlite3)"
-    } finally {
-        Pop-Location
-        Remove-Item $tempScript -Force -ErrorAction SilentlyContinue
-    }
-}
-
-if ($Verbose -and (Test-Path $DatabasePath)) {
-    $dbSize = (Get-Item $DatabasePath).Length / 1KB
-    Write-Info "Database size: $([math]::Round($dbSize, 2)) KB"
-}
-
-# ============================================================================
-# Step 3: Verification
-# ============================================================================
-if (-not $SkipVerify) {
-    Write-Step "3" $totalSteps "Verifying database..."
-
-    # Expected table counts
-    $expectedTables = @{
-        "test_products" = 16
-        "test_orders" = 20
-        "test_jsonb_docs" = 6
-        "test_articles" = 8
-        "test_users" = 9
-        "test_measurements" = 200
-        "test_embeddings" = 20
-        "test_locations" = 15
-        "test_categories" = 17
-        "test_events" = 100
-    }
-
-    # Known non-seed tables that are expected to exist (SQLite internals,
-    # SpatiaLite system tables/views, FTS virtual tables created by seed SQL, etc.)
-    $knownSystemTables = @(
-        # SQLite internals
-        "sqlite_sequence", "sqlite_stat1", "sqlite_stat4",
-        # SpatiaLite tables
-        "ElementaryGeometries", "KNN2", "SpatialIndex",
-        "data_licenses", "geometry_columns", "geometry_columns_auth",
-        "geometry_columns_field_infos", "geometry_columns_statistics",
-        "geometry_columns_time", "spatial_ref_sys", "spatial_ref_sys_aux",
-        "spatialite_history", "sql_statements_log",
-        "views_geometry_columns", "views_geometry_columns_auth",
-        "views_geometry_columns_field_infos", "views_geometry_columns_statistics",
-        "virts_geometry_columns", "virts_geometry_columns_auth",
-        "virts_geometry_columns_field_infos", "virts_geometry_columns_statistics",
-        # SpatiaLite views
-        "geom_cols_ref_sys", "spatial_ref_sys_all",
-        "vector_layers", "vector_layers_auth",
-        "vector_layers_field_infos", "vector_layers_statistics",
-        # FTS5 shadow tables (from seed SQL)
-        "test_articles_fts", "test_articles_fts_config", "test_articles_fts_content",
-        "test_articles_fts_data", "test_articles_fts_docsize", "test_articles_fts_idx",
-        # Migration tracking (created by seed SQL)
-        "_mcp_migrations"
-    )
-
-    $verifyScript = @"
-import Database from 'better-sqlite3';
-const db = new Database(process.argv[2], { readonly: true });
-
-const tables = db.prepare("SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'test_%'").all();
-console.log('TABLES:' + tables.map(t => t.name).join(','));
-
-for (const table of tables) {
-    const count = db.prepare('SELECT COUNT(*) as cnt FROM ' + table.name).get();
-    console.log('COUNT:' + table.name + ':' + count.cnt);
-}
-
-db.close();
-"@
-
-    $dbMcpRoot = Split-Path -Parent $ScriptDir
-    # Write temp script to project directory so node can resolve dependencies
-    $tempVerify = Join-Path $dbMcpRoot ".verify-test-db.js"
-    $verifyScript | Out-File -FilePath $tempVerify -Encoding utf8 -NoNewline
-
-    Push-Location $dbMcpRoot
-    try {
-        $output = node $tempVerify $DatabasePath 2>&1
-
-        $tableList = ($output | Where-Object { $_ -match "^TABLES:" }) -replace "^TABLES:", ""
-        $tables = $tableList -split ","
-
-        Write-Host "`n  Table verification:" -ForegroundColor Yellow
-
-        $allPassed = $true
-        foreach ($line in $output | Where-Object { $_ -match "^COUNT:" }) {
-            $parts = ($line -replace "^COUNT:", "") -split ":"
-            $tableName = $parts[0]
-            $actualCount = [int]$parts[1]
-
-            if ($expectedTables.ContainsKey($tableName)) {
-                $expectedCount = $expectedTables[$tableName]
-                if ($actualCount -eq $expectedCount) {
-                    Write-Host "    [pass] " -ForegroundColor Green -NoNewline
-                    Write-Host "$tableName" -NoNewline
-                    Write-Host " ($($actualCount) rows)" -ForegroundColor Gray
-                } else {
-                    Write-Host "    [fail] " -ForegroundColor Red -NoNewline
-                    Write-Host "$tableName" -NoNewline
-                    Write-Host " (expected $($expectedCount), got $($actualCount))" -ForegroundColor Red
-                    $allPassed = $false
-                }
-            }
-        }
-
-        if ($allPassed) {
-            Write-Success "All tables verified successfully"
-        } else {
-            Write-Warn "Some tables have unexpected row counts"
-        }
-
-        # Check for unexpected non-seed tables (test artifacts left behind)
-        Write-Host "`n  Artifact check:" -ForegroundColor Yellow
-        $allTableNames = ($output | Where-Object { $_ -match "^TABLES:" }) -replace "^TABLES:", ""
-        # Get ALL tables from sqlite_master, not just test_* ones
-        $allTablesScript = @"
-import Database from 'better-sqlite3';
-const db = new Database(process.argv[2], { readonly: true });
-const all = db.prepare("SELECT name, type FROM sqlite_master WHERE type IN ('table','view') ORDER BY name").all();
-for (const t of all) console.log('ALL:' + t.name);
-db.close();
-"@
-        $tempAllTables = Join-Path $dbMcpRoot ".check-artifacts.js"
-        $allTablesScript | Out-File -FilePath $tempAllTables -Encoding utf8 -NoNewline
-        $allOutput = node $tempAllTables $DatabasePath 2>&1
-        Remove-Item $tempAllTables -Force -ErrorAction SilentlyContinue
-
-        $unexpectedTables = @()
-        foreach ($line in $allOutput | Where-Object { $_ -match "^ALL:" }) {
-            $name = ($line -replace "^ALL:", "").Trim()
-            $isExpected = $expectedTables.ContainsKey($name) -or $knownSystemTables -contains $name
-            if (-not $isExpected) {
-                $unexpectedTables += $name
-            }
-        }
-
-        if ($unexpectedTables.Count -gt 0) {
-            Write-Warn "Found $($unexpectedTables.Count) unexpected table(s) — possible stale test artifacts:"
-            foreach ($ut in $unexpectedTables) {
-                Write-Host "    [stale] " -ForegroundColor Yellow -NoNewline
-                Write-Host $ut -ForegroundColor Gray
-            }
-            Write-Info "These may be R-Tree shadow tables, _mcp_migrations, temp_* leftovers, or FTS tables from testing."
-            Write-Info "Consider running a full reset to remove them."
-        } else {
-            Write-Success "No stale test artifacts found"
-        }
-
-    } finally {
-        Pop-Location
-        Remove-Item $tempVerify -Force -ErrorAction SilentlyContinue
-    }
-}
-
-Write-Host "`n========================================================" -ForegroundColor Green
-Write-Host "                    Reset Complete!                      " -ForegroundColor Green
-Write-Host "========================================================" -ForegroundColor Green

package/test-server/sample.csv

@@ -1,11 +0,0 @@
-id,name,category,price,quantity,created_at
-1,Widget A,Electronics,29.99,100,2026-01-15
-2,Widget B,Electronics,49.99,75,2026-01-16
-3,Gadget X,Home,19.99,200,2026-01-17
-4,Gadget Y,Home,24.99,150,2026-01-18
-5,Tool Alpha,Industrial,99.99,50,2026-01-19
-6,Tool Beta,Industrial,149.99,30,2026-01-20
-7,Component A,Electronics,9.99,500,2026-01-21
-8,Component B,Electronics,14.99,400,2026-01-22
-9,Fixture X,Home,39.99,80,2026-01-23
-10,Fixture Y,Home,59.99,60,2026-01-24

package/test-server/test-agent-experience.md

@@ -1,243 +0,0 @@
-# Agent Experience Test — db-mcp (SQLite)
-
-> **Purpose:** Validate that the slim `instructions` field + `sqlite://help` resources are sufficient for an agent to operate the server cold — with **zero** schema info, tool hints, or checklists in the prompt.
-
-## How to Run
-
-Run **each pass** as a separate conversation with the corresponding `--tool-filter`. Each pass tests whether the agent can complete realistic tasks using only the tools + help resources available under that filter.
-
-| Pass   | `--tool-filter`                | Tools                                | Scenarios |
-| ------ | ------------------------------ | ------------------------------------ | --------- |
-| Pass 1 | `starter`                      | Core, JSON, Text (~50)               | 1–12      |
-| Pass 2 | `analytics`                    | Core, JSON, Stats (~52)              | 13–16     |
-| Pass 3 | `search`                       | Core, Text, Vector (~38)             | 17–18     |
-| Pass 4 | `codemode`                     | Code Mode only (1+3)                 | 19–20     |
-| Pass 5 | `core,geo`                     | Core, Geo (~22)                      | 21–23     |
-| Pass 6 | `core,admin`                   | Core, Admin (~44)                    | 24–26     |
-| Pass 7 | `core,introspection,migration` | Core, Introspection, Migration (~27) | 27–29     |
-
-> **Important:** Do NOT combine passes. Each pass is a fresh conversation with a clean context. The agent has never seen this database before.
-
-## Rules
-
-1. **Do NOT read** `test-tools.md`, `test-group-tools.md`, or any other test documentation before running these scenarios
-2. **Do NOT read** source code files (`src/`) — you are a user, not a developer
-3. **DO** use the MCP instructions you received during initialization + `sqlite://help` resources
-4. **DO** discover the database schema via `sqlite://schema` or `sqlite://tables` resources
-5. **DO** read group-specific help (`sqlite://help/{group}`) when you need reference for unfamiliar tools
-6. Use the test database (`test-server/test.db`) which is already connected
-
-## Success Criteria
-
-| Symbol | Meaning                                                               |
-| ------ | --------------------------------------------------------------------- |
-| ✅     | Agent completed the task correctly without external help              |
-| ⚠️     | Agent completed but needed multiple retries or used wrong tools first |
-| ❌     | Agent failed or produced incorrect results                            |
-| 📖     | Agent had to read help resources — note which ones                    |
-
-Track **every** help resource read and whether it provided what was needed. Gaps are the actionable finding.
-
-## Reporting Format
-
-For each scenario, report:
-
-```
-### Scenario N: [title]
-**Result:** ✅/⚠️/❌
-**Resources read:** sqlite://help, sqlite://help/json (or "none beyond instructions")
-**Tools used:** sqlite_read_query, sqlite_json_query, ...
-**Issues:** (any gaps in help content, confusing tool names, missing examples)
-```
-
----
-
-## Pass 1: `starter`
-
-**Tool groups under test:** `core` (10), `json` (24), `text` (14), `codemode` (1)
-
-### Phase 1 — Discovery
-
-#### Scenario 1 — What's in this database?
-
-List all tables and briefly describe what each one contains.
-
-#### Scenario 2 — Table deep dive
-
-Pick the most interesting table and fully characterize it: row count, column types, sample data, indexes, and any foreign key relationships.
-
-#### Scenario 3 — Health check
-
-Is the database healthy? What backend and mode is it running in? What are the key PRAGMA settings?
-
-### Phase 2 — Core Operations
-
-#### Scenario 4 — Filtered read
-
-Find all products in the "electronics" category priced above $50, sorted by price descending.
-
-#### Scenario 5 — Aggregation
-
-What is the total revenue (sum of total_price) per order status? Which status has the highest revenue?
-
-#### Scenario 6 — Write and verify
-
-Create a new product called "Test Widget" in the "gadgets" category priced at $29.99, then verify it was inserted. Clean up after.
-
-### Phase 3 — JSON Operations
-
-#### Scenario 7 — JSON extraction
-
-Extract the `name` and `version` fields from the JSON documents in `test_jsonb_docs`. Which documents have a version above 2?
-
-#### Scenario 8 — Nested JSON
-
-Find the deeply nested value at `$.nested.level1.level2` in `test_jsonb_docs`. Which row has it?
-
-#### Scenario 9 — JSON analysis
-
-Analyze the JSON schema of the `doc` column in `test_jsonb_docs`. What field types and nesting patterns exist?
-
-### Phase 4 — Text & Search
-
-#### Scenario 10 — Full-text search
-
-Search the articles for content about "database" and "search". Rank results by relevance.
-
-> **Note:** FTS5 tables may need to be created/rebuilt before searching. The agent should figure this out from help resources.
-
-#### Scenario 11 — Fuzzy matching
-
-Find users whose names are similar to "Jon" (fuzzy match). What did you find?
-
-#### Scenario 12 — Text processing
-
-Normalize and validate the email addresses in `test_users`. Are any malformed?
-
----
-
-## Pass 2: `analytics`
-
-**Tool groups under test:** `core` (10), `json` (24), `stats` (14), `codemode` (1)
-
-### Phase 5 — Statistics
-
-#### Scenario 13 — Descriptive stats
-
-Compute descriptive statistics (mean, median, std dev, min, max) for the `temperature` column in `test_measurements`. Break it down by `sensor_id`.
-
-#### Scenario 14 — Outlier detection
-
-Are there any outlier temperature readings in `test_measurements`? What method did you use?
-
-#### Scenario 15 — Correlation
-
-Is there a correlation between temperature and humidity in `test_measurements`? How strong?
-
-#### Scenario 16 — Cross-table analysis
-
-Which products have the most orders? Join the data and present a ranked summary with product name, order count, and total revenue.
-
----
-
-## Pass 3: `search`
-
-**Tool groups under test:** `core` (10), `text` (14), `vector` (12), `codemode` (1)
-
-### Phase 6 — Vector & Semantic Search
-
-#### Scenario 17 — Similarity search
-
-Find the 3 embeddings most similar to the first embedding in `test_embeddings`. What categories are they?
-
-#### Scenario 18 — Hybrid search
-
-Search articles for "database" using full-text search, then find semantically similar embeddings. Can the agent combine both approaches?
-
----
-
-## Pass 4: `codemode`
-
-**Tool groups under test:** `codemode` (1) + built-in resources (3)
-
-### Phase 7 — Code Mode Discovery & Efficiency
-
-#### Scenario 19 — Cold-start Code Mode
-
-Using only `sqlite_execute_code`, list all tables, pick one, and run a query against it. Can the agent discover the `sqlite.*` API without external help?
-
-#### Scenario 20 — Multi-step workflow
-
-Using only `sqlite_execute_code`, find the top 5 products by order count with total revenue — in a single code execution. Compare the token efficiency vs what Pass 1/2 would require.
-
----
-
-## Pass 5: `core,geo`
-
-**Tool groups under test:** `core` (10), `geo` (5), `codemode` (1)
-
-### Phase 8 — Geospatial
-
-#### Scenario 21 — Distance between cities
-
-What is the distance between New York and Paris based on the coordinates in `test_locations`?
-
-#### Scenario 22 — Nearby locations
-
-Find all locations within 100km of London. How many are there?
-
-#### Scenario 23 — Bounding box
-
-Find all locations within the bounding box covering Western Europe (lat 40–55, lon -5 to 15).
-
----
-
-## Pass 6: `core,admin`
-
-**Tool groups under test:** `core` (10), `admin` (27), `codemode` (1)
-
-### Phase 9 — Admin & Data Quality
-
-#### Scenario 24 — Data quality audit
-
-Audit the database for data quality issues: NULL values, duplicate entries, orphaned foreign keys, unused indexes. Summarize findings.
-
-#### Scenario 25 — Database health & optimization
-
-Run an integrity check, analyze the database, then suggest and apply optimizations.
-
-#### Scenario 26 — Backup and restore
-
-Create a backup of the database. Verify the backup exists and is valid.
-
----
-
-## Pass 7: `core,introspection,migration`
-
-**Tool groups under test:** `core` (10), `introspection` (10), `migration` (7), `codemode` (1)
-
-### Phase 10 — Schema Analysis & Migration
-
-#### Scenario 27 — Dependency graph
-
-Map out the foreign key dependency graph. Which tables have the most dependencies? What's the topological sort order?
-
-#### Scenario 28 — Cascade simulation
-
-What would happen if `test_products` row 1 were deleted? Simulate the cascade impact on related tables.
-
-#### Scenario 29 — Migration workflow
-
-Initialize migration tracking, then create and apply a migration that adds a `description` column to `test_products`. Roll it back after verifying.
-
----
-
-## Post-Test Summary
-
-Compile findings across all passes into:
-
-1. **Help resource gaps** — scenarios where help content was missing, incomplete, or misleading
-2. **Discovery friction** — cases where the agent struggled to find the right tool or resource
-3. **Suggested improvements** — specific additions to `src/constants/server-instructions/*.md`
-
-> **Key metric:** How many of the 29 scenarios did the agent complete on the first try with ≤1 help resource read? This measures whether the instructions + tool descriptions are self-sufficient.