Migrationpilot

MigrationPilot

[npm version] [npm downloads] [CI] [Node] [VS Code] [License: MIT]

Block unsafe Postgres migrations before merge.

Local, deterministic analysis for PostgreSQL migrations. Uses PostgreSQL's parser, checks 112 rules, and exits non-zero in CI. No account required. MIT.

npx migrationpilot analyze migration.sql

Try it in your browser · GitHub Action · Documentation

Benchmark

56 labelled files. Author-built corpus. Tools pinned.

Methodology · Corpus · What MigrationPilot missed · Reproduce: pnpm build && node bench/run.mjs

A finding

-- migration.sqlALTER TABLE users ADD CONSTRAINT users_email_unique UNIQUE (email);

$ migrationpilot analyze migration.sql
  ✗ MigrationPilot —  RED  Score: 80/100  migration.sql  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  ─  1 statement · 2 critical · rollback GREEN
┌─────┬─────────────────────────────────────────────┬─────────────────────────┬────────┬────────────┐│ #   │ Statement                                   │ Lock Type               │ Risk   │ Long lock? │├─────┼─────────────────────────────────────────────┼─────────────────────────┼────────┼────────────┤│ 1   │ ALTER TABLE users ADD CONSTRAINT users_e... │ ACCESS EXCLUSIVE        │  RED   │ YES        │└─────┴─────────────────────────────────────────────┴─────────────────────────┴────────┴────────────┘
  Violations:
  ✗ [MP004] CRITICAL (line 1)    DDL statement acquires ACCESS EXCLUSIVE lock without a preceding SET lock_timeout. Without a timeout, this statement could block the lock queue indefinitely if it can't acquire the lock, causing cascading query failures.
    Safe alternative:    -- Set a timeout so DDL fails fast instead of blocking the queue    SET lock_timeout = '5s';    ALTER TABLE users ADD CONSTRAINT users_email_unique UNIQUE (email)    RESET lock_timeout;
    Why: Without lock_timeout, if the table is locked by another query, your DDL waits indefinitely. All subsequent queries pile up behind it in the lock queue, causing cascading timeouts across your application. GoCardless enforces a 750ms lock_timeout for this reason.    Docs: https://migrationpilot.dev/rules/mp004
  ✗ [MP027] CRITICAL (line 1)    Adding UNIQUE constraint "users_email_unique" on "users" scans the entire table under ACCESS EXCLUSIVE lock. Create the index concurrently first, then use USING INDEX.
    Safe alternative:    -- Step 1: Create the unique index concurrently (non-blocking)    CREATE UNIQUE INDEX CONCURRENTLY users_email_unique_idx ON users (...);
    -- Step 2: Add the constraint using the pre-built index (instant)    ALTER TABLE users ADD CONSTRAINT users_email_unique UNIQUE USING INDEX users_email_unique_idx;
    Why: ALTER TABLE ADD CONSTRAINT UNIQUE builds a unique index while holding ACCESS EXCLUSIVE lock, blocking all reads and writes for the entire scan. Instead, create the unique index concurrently (non-blocking), then attach it as a constraint with USING INDEX.    Docs: https://migrationpilot.dev/rules/mp027
  Risk Factors:    Lock Severity        ██████████ 40/40 — ACCESS EXCLUSIVE (long-held)    Rule Violations      ████████░░ 80/100 — 2 critical
  112 rules checked in 11ms

Exit code is 2. The Risk column combines what a statement's lock does with what the rules found in it, so a statement carrying a critical violation reads RED whatever its lock costs. The lock half of that is capped without a database connection — table size and query frequency need one. See Production context.

Contents

Install · AI coding agents · CI · What it checks · Beyond one file · Configuration · Output · Production context · Comparison · Pricing · Architecture · API

Install

npx migrationpilot analyze migration.sql   # no installnpm install -g migrationpilot              # global

Node 22 or newer. The PostgreSQL parser ships compiled in, so there is nothing else to set up. Exit codes are the same everywhere: 0 clean, 1 warnings under --fail-on warning, 2 critical.

Packaged builds land with each release, including single-file executables for Linux, macOS and Windows on the release page for machines without Node. The Windows .exe is not code-signed, so SmartScreen and most browsers will warn about it on download — SHA256SUMS on the same release is how you check you got the file we published, not a signature.

brew install mickelsamuel/migrationpilot/migrationpilotdocker run --rm -v "$PWD:/work" ghcr.io/mickelsamuel/migrationpilot:1 analyze migration.sql

On Windows in Git Bash, MSYS rewrites paths inside the mount flag, so use the Windows-form working directory instead:

docker run --rm -v "$(pwd -W):/work" ghcr.io/mickelsamuel/migrationpilot:1 analyze migration.sql

AI coding agents

Agents write migrations now. They are good at SQL and bad at knowing which statement takes an ACCESS EXCLUSIVE lock on a table with 40 million rows, and by then the outage has already happened.

MCP server. Seven tools, the important one being check_before_apply: a pass/fail gate the agent calls before it writes or runs DDL. It resolves your .migrationpilotrc.yml exactly like the CLI does, so its verdict is the verdict CI will give.

{  "mcpServers": {    "migrationpilot": { "command": "npx", "args": ["migrationpilot-mcp"] }  }}

Claude Code plugin. integrations/claude-code/ pairs a skill that tells Claude to check migrations with a PreToolUse hook that blocks the tool call when it doesn't. It fails open on purpose: a missing install, unparseable SQL, or a timeout lets the call through with a note on stderr, because a guardrail that breaks your workflow when it can't run gets uninstalled.

claude plugin install ./integrations/claude-code

Cursor and Copilot. Copy integrations/cursor/migrationpilot.mdc into .cursor/rules/, or paste integrations/copilot/copilot-instructions-snippet.md into .github/copilot-instructions.md. Both tell the agent when to run MigrationPilot and that suppressing a rule to get past a violation is the user's call, not the agent's.

CI

GitHub Action

# .github/workflows/migration-check.ymlname: Migration Safety Checkon: [pull_request]
# New repositories default the workflow token to read-only; the report comment# needs pull-request write.permissions:  contents: read  pull-requests: write
jobs:  check:    runs-on: ubuntu-latest    steps:      - uses: actions/checkout@v4      - uses: mickelsamuel/migrationpilot@v1        with:          migration-path: "migrations/*.sql"          fail-on: critical

Posts a report as a PR comment, fails the check on critical violations, and writes a SARIF file. To feed it into Code Scanning, add an upload step (needs Advanced Security on private repos):

      - uses: github/codeql-action/upload-sarif@v3        if: always()        with:          sarif_file: migrationpilot-results.sarif

Without the permissions block the Action still runs. It warns, analyzes every file matching the glob instead of only the ones the PR changed, and skips the comment. The check verdict, the SARIF file and the inline annotations come from the analysis either way.

A migration the Action can't read or parse fails the check, since nothing analyzed it, and the failure names the file and the parser's error. Files that aren't plain SQL, like ones with psql meta-commands or template placeholders, go under ignore in the config. With fail-on: never they're reported and the check passes.

The Action reads the same config file as the CLI, so ignore, rules, failOn, pgVersion and extends all apply there. An input you set in the workflow wins over the config. With an Org license key, so do the config's policy block and audit log, and the policy wins over the inputs: exclude can't drop a required rule, and a blocked statement fails the check even under fail-on: never.

Outputs: risk-level, violations, sarif-file.

Pre-commit

migrationpilot hook install writes a plain git hook and is Husky-aware. With the pre-commit framework instead:

repos:  - repo: https://github.com/mickelsamuel/migrationpilot    rev: v1.6.0    hooks:      - id: migrationpilot        args: [--fail-on, warning]

Clean files print nothing. Only migrations with violations are reported.

If pre-commit install answers Cowardly refusing to install hooks with 'core.hooksPath' set, something else already owns your hooks directory — Husky sets it. Check with git config core.hooksPath, then either git config --unset-all core.hooksPath and let pre-commit manage the hooks, or keep Husky and run migrationpilot hook install, which appends to .husky/pre-commit instead of fighting it.

GitLab CI

include:  - remote: 'https://raw.githubusercontent.com/mickelsamuel/migrationpilot/v1.6.0/integrations/gitlab/.gitlab-ci-migrationpilot.yml'
migrationpilot:  variables:    MIGRATIONPILOT_PATH: db/migrate

Runs on merge requests that touch migrations, keeps the JSON report as an artifact, and annotates the MR diff through GitLab Code Quality.

What it checks

112 rules: 34 critical, 78 warning, 20 auto-fixable with --fix. Ten that matter most:

Browse all 112 rules, or run migrationpilot explain MP027 for one. The handbook is 20 chapters on why each hazard bites and what to do instead.

Rules adapt to --pg-version (9 through 18): REINDEX CONCURRENTLY from 12, DETACH PARTITION CONCURRENTLY from 14, the native NOT NULL ... NOT VALID path from 18.

Beyond one file

analyze --fix rewrites the 20 fixable violations in place. The rest of the surface:

Twenty-four commands in total. migrationpilot --help lists them.

Sequence analysis is what a per-file linter cannot see. Three migrations that each look fine can still take one table down together:

$ migrationpilot check migrations/
  ⚠ [SQ001] WARNING cumulative-lock-budget    "orders" is locked for an estimated 2m across 2 statements in 2 files — over the 1m budget for one deploy.  ⚠ [SQ002] WARNING hot-table-multi-touch    "orders" is locked by 3 files in this sequence. Each one queues behind live traffic on its own — fold them into one migration so the table takes the hit once.

Tune it with --lock-budget <seconds> and --hot-table-threshold <files>, turn it off with --no-sequence, and make it blocking with --fail-on-sequence.

--fail-on irreversible is stricter than critical: it also blocks migrations that destroy data with no down file.

Configuration

Zero-config is the default. check with no directory detects your framework, finds its migrations, and analyzes them in apply order. Fourteen are supported: Flyway, Liquibase, Alembic, Django, Knex, Prisma, TypeORM, Drizzle, Sequelize, goose, dbmate, Sqitch, Rails, Ecto. Force one with --framework prisma, or pipe any generator through --from-command:

migrationpilot check --from-command "python manage.py sqlmigrate myapp 0042"

# .migrationpilotrc.ymlextends: "migrationpilot:strict"pgVersion: 16failOn: warningrules:  MP037: false                 # off  MP004: { severity: warning } # downgrade  MP013: { threshold: 5000 }   # retuneignore:  - "migrations/seed_*.sql"

Five presets: recommended (default), strict, ci, startup, enterprise. Inline, -- migrationpilot-disable MP001 suppresses a rule for the next statement and -- migrationpilot-disable-file MP001 does it for the whole file. Name no rule and it suppresses all of them.

Org policy

With an Org plan key in MIGRATIONPILOT_LICENSE_KEY (or --license-key, or the Action's license-key input), analyze, check, precommit and the Action apply the config's policy block. watch, plan, simulate and mutation-test keep the required rules on and hold the floor too, but only the gates fail on a blocked pattern.

policy:  requiredRules: [MP001, MP004]   # run whatever rules:, a preset, --exclude or a disable comment says  severityFloor: critical         # a critical rule can't be set below critical  blockedPatterns: ["DROP TABLE", "TRUNCATE"]   # fail the run whatever --fail-on says  requireReviewPatterns: ["ALTER TYPE"]         # reported for team reviewauditLog:  enabled: true                   # one JSONL line per check

Each override prints a note saying what it kept on. A rule that ships as a warning stays a warning under a critical floor. Patterns match whole words, ignoring case, whitespace and comments. Without a valid key, the run says the policy isn't applied rather than passing quietly. --no-config skips the config file and the policy with it, so enforce from CI, where the config gets reviewed like code. The MCP server doesn't read the config.

Ed25519 license keys validate client-side, with no network call. --offline skips update checks and every other network call. There is no telemetry.

Output

--format text (default), json, sarif, or markdown, plus --quiet for one gcc-style line per violation and --verbose for per-statement pass/fail.

{  "$schema": "https://migrationpilot.dev/schemas/report-v1.json",  "version": "1.6.0",  "file": "migrations/001.sql",  "riskLevel": "RED",  "riskScore": 80,  "violations": []}

SARIF feeds GitHub Code Scanning, VS Code and IntelliJ: migrationpilot analyze migration.sql --format sarif --output results.sarif.

Production context

Pass --database-url and MigrationPilot opens one read-only connection to read pg_class, pg_stat_statements and pg_stat_activity. It reads no user data and runs no DDL.

That turns risk scoring from a guess into a measurement, and gives three rules the numbers they have nothing to say without: MP013 (DDL on a high-traffic table), MP014 (long-held locks on a table with millions of rows), MP019 (ACCESS EXCLUSIVE while connections are piling up).

GREEN is 0-24, YELLOW 25-49, RED 50-100.

Comparison

Squawk: 40 rules as of v2.62.0 (Aug 2026). Atlas gates migrate lint behind a Pro login in the official binary since v0.38 (Oct 2025); the Community build keeps a basic analyzer set, but the PostgreSQL lock analyzers are Pro-only. It could not be benchmarked without a paid account. The methodology records the exact command and its refusal.

Pricing

Everything the linter does is free and unmetered: all 112 rules including the production-context ones, auto-fix, sequence analysis, simulate, every output format, the GitHub Action, the MCP server. No account, no seat count, no telemetry, MIT.

The $499/year Org plan turns the free linter into an enforceable control: one policy across repositories that developers cannot quietly disable, a JSONL audit trail of every check, and direct support from the maintainer.

Org plan · Full pricing

Architecture

src/├── parser/ locks/         # libpg-query WASM, lock classification├── rules/ fixer/          # 112 rules and the 20-rule auto-fixer├── analysis/ scoring/     # shared pipeline, transaction boundaries, risk 0-100├── sequence/ lockqueue/   # cross-file SQ rules, lock queue modelling├── simulate/ mutate/      # PGlite execution, mutation-testing operators├── cascade/ graph/ schema/ prediction/ templates/├── production/ frameworks/ plugins/ output/ generator/├── mcp/ action/ config/ hooks/ watch/ drift/ history/├── policy/ auth/ license/ team/ audit/ billing/ usage/ doctor/├── index.ts               # programmatic API, 69 value exports plus types└── cli.ts                 # 24 commands

Programmatic API

import { analyzeSQL, allRules, parseMigration, classifyLock } from 'migrationpilot';
const result = await analyzeSQL(sql, 'migration.sql', 17, allRules);console.log(result.violations, result.overallRisk);

Sixty-nine value exports plus full TypeScript types. allRules is the same rule set the CLI runs.

Development

pnpm installpnpm test        # 1945 tests across 72 filespnpm build       # CLI 1.4MB, Action 1.7MB, API 639KB, MCP 1.7MBpnpm lint && pnpm typecheckpnpm dev analyze path/to/migration.sql

CONTRIBUTING.md · SECURITY.md · CHANGELOG.md

License

MIT