encore-migrate（再次迁移）

Migrate to Encore

This skill guides migrating any existing backend application to Encore, one migration unit at a time. It supports any source language or framework and targets both Encore.ts and Encore Go. A migration-plan.md summary file and migration-plan/ directory of per-unit detail files are created at the Encore project root to track progress across sessions. This skill contains no Encore code examples — it delegates all Encore-specific implementation to the appropriate language-specific skills.

Phase Detection

Before doing anything, determine which phase to enter:

• No migration-plan.md exists in the Encore project directory → Start at Phase 1: DISCOVER
• migration-plan.md exists but no migration-plan/ directory → Resume at Phase 2: PLAN (discovery done, detail files not yet written)
• migration-plan/ directory exists with pending units (any unit in the summary with status pending or in progress) → Resume at Phase 3: MIGRATE
• All units in the summary are migrated, skipped, or manual validation needed → Go to Phase 4: COMPLETE

Resuming a Migration (Phase 3)

When migration-plan.md and migration-plan/ exist with pending units:

1. Read migration-plan.md (summary only — do NOT read all detail files)
2. Report current status to the user — for example: "3 of 7 units migrated, next suggested: billing (all its dependencies are migrated)"
3. Ask the user what they would like to work on next, offering a suggestion based on the dependency order in the plan

Phase 1 — Discover

1. Gather Information

Ask the user for:

• Path to the source system (the existing codebase being migrated)
• Local URL where the source system runs (if applicable — needed for HTTP comparison validation later)
• Target language: Encore.ts or Encore Go

2. Analyze the Source Codebase

Read the source codebase and inventory all entities:

3. Identify Frontend Code

Full-stack repos and monorepos often mix backend and frontend code. The migration targets backend only — frontend code is out of scope.

Detect frontend directories and mark them as out of scope. Common indicators:

**Flag framework server-side code that *should* be migrated.** Some frontend frameworks embed backend logic that contains API endpoints, database queries, or server-side business logic:

When framework server-side code is found, ask the user what to do with it. Not all server-side code should move to Encore — sometimes a thin backend layer (BFF, auth proxy, SSR data fetching) should stay in the frontend framework alongside an Encore backend.

Present the user with what was found and ask:

"I found server-side routes in your app (e.g., pages/api/users.ts, app/billing/route.ts). These contain backend logic that *could* be migrated to Encore, but some teams prefer to keep a thin server layer in their frontend framework for things like SSR data fetching or BFF proxying. Would you like to: 1. Migrate all server-side routes to Encore 2. Migrate some — I'll list them and you pick which ones move 3. Keep all in — only migrate the standalone backend code"

Based on the user's choice:

• Migrate all: Extract the backend logic into migration units. Leave frontend rendering code out of scope. Note in the migration plan which source files contain mixed frontend/backend code.
• Migrate some: Present the list of server-side routes and let the user select. Include selected routes in migration units, mark the rest as out of scope.
• Keep all: Mark all framework server-side code as out of scope alongside the frontend. Only standalone backend code (Express routes, standalone API servers, etc.) enters migration units.

Report to the user: List all detected frontend directories and the decision made about framework server-side code. Example: "I found a Next.js frontend in app/ — the React components are out of scope. You chose to migrate 8 of the 12 API routes from pages/api/ to Encore and keep 4 thin proxy routes in Next.js."

4. Group Entities into Migration Units

Group the discovered entities into migration units using these heuristics in priority order:

1. Existing service boundaries — If the source app already has services, modules, or packages, use those as the starting point for chunks
2. URL path prefixes — Group endpoints sharing a path prefix (e.g., /users/*, /billing/*)
3. Shared database tables — Endpoints that read/write the same tables belong together
4. Shared types/models — Endpoints that share request/response types or domain models

Chunk sizing: Aim for 5-15 endpoints per migration unit. If a group exceeds ~15 endpoints, suggest splitting it further (e.g., users-crud and users-admin). If a group has fewer than 3 endpoints, consider merging it with a related chunk.

Cross-cutting concerns get their own migration units: auth, secrets, and standalone infrastructure (pub/sub topics, cron jobs not tightly coupled to one service) are separate units since they follow different dependency tiers.

For monoliths with no clear boundaries: Fall back to URL path prefix grouping, then ask: "These groupings are based on URL paths — would you like to reorganize them by domain?"

5. Present the Migration Units

Present the migration units to the user as a summary table:

Include total counts (e.g., "7 migration units covering 42 endpoints, 3 databases"). For each unit, assess overall migration complexity:

• Low — direct Encore equivalents exist, straightforward mapping
• Medium — requires restructuring or has partial Encore equivalents
• High — no direct equivalent, needs redesign or custom solution

Offer to show the detail of any unit if the user wants to inspect what's inside before confirming.

6. Show Code Previews

For 2-3 representative entities (pick a mix of simple and complex from different units), show a short "before and after" preview of what the source code looks like now and what the Encore version will look like. Use the appropriate language-specific skill to inform the preview. Keep previews brief — one endpoint, one query, or one topic declaration is enough per preview.

7. Confirm with the User

Ask the user to confirm the migration units are correct. Specifically ask:

• "Are there any services, endpoints, or other entities I missed?"
• "Would you like to split, merge, or rename any of these migration units?"
• "Is there anything you want to exclude from the migration?"

8. Iterate if Needed

If the user identifies missing entities or wants to adjust chunk boundaries, update the units and re-present the summary table. Repeat until the user confirms the migration units are accurate.

Phase 2 — Plan

1. Check for Existing Encore Project

Check if an Encore project already exists at the target path (look for encore.app file). If yes, confirm with the user that this is the correct project. If no, help create one by invoking the encore-getting-started skill (or encore-go-getting-started for Go).

2. Gather Target Information

Ask the user for:

• Path to the Encore project (where the migrated code will live)
• Local URL where the Encore app will run (default: http://localhost:4000)

3. Determine Dependency Order

Order migration units based on dependencies. Follow this tier order:

1. Secrets / config — no dependencies, needed by everything
2. Databases — schema and migrations must exist before services can use them
3. Auth — auth handlers are needed before protected endpoints
4. Leaf units — units with no cross-service dependencies
5. Dependent units — units that depend on already-migrated units
6. Pub/Sub topics and subscriptions — often depend on services being in place
7. Cron jobs — typically depend on service endpoints

Within each tier, suggest the simplest unit first (fewest endpoints, smallest schema, least complexity).

4. Write migration-plan.md (Summary)

Write the migration-plan.md summary file to the Encore project root using the template in the "migration-plan.md Format" section below. Fill in all migration units with status pending.

5. Write Detail Files

Create a migration-plan/ directory at the Encore project root. Write one detail file per migration unit using the template in the "Detail File Format" section below. Each file is named migration-plan/<unit-name>.md.

6. Propose First Unit

Propose the first migration unit, explaining why it should go first based on the dependency order. Wait for user approval before proceeding to Phase 3.

Phase 3 — Migrate (Loop)

1. Identify Next Unit

Read migration-plan.md (summary only) and identify the next pending migration unit based on the dependency order.

2. Suggest and Confirm

Suggest the next unit to migrate and explain why this one is next (e.g., "This unit has no dependencies on unmigrated units" or "The database must exist before we can migrate the service that uses it"). Ask the user if they want to proceed with this unit or pick a different one.

3. Load the Unit Detail

Read the detail file for the chosen unit (migration-plan/<unit-name>.md). Do NOT read detail files for other units.

4. Migrate Each Entity

For each entity in the unit:

a. Implement

Invoke the appropriate language-specific skill based on the entity type and target language:

b. Migrate Tests

If the source entity has associated tests, migrate them using the appropriate testing skill (encore-testing or encore-go-testing). Adapt test assertions to match Encore API patterns. If the source entity has no tests, note this in the detail file.

c. Validate

Three validation layers are applied to each entity before it can be marked as migrated. Every entity must go through all applicable layers.

Layer 1: Test Migration (Primary)

• When migrating an entity, also migrate its associated tests
• Use the encore-testing skill (or encore-go-testing for Go) to implement the tests
• Run the tests — they must pass before the entity can be marked as migrated
• If the source entity had no tests, note "no source tests" in the plan and rely on the other layers

Layer 2: HTTP Comparison (Endpoints Only, Best-Effort)

When both systems are running locally, call the same endpoint on both the source system and the Encore app, then compare:

• HTTP status code — must match
• Response body structure — keys and shape must match (values may differ for dynamic data like timestamps or IDs)

Skip this layer when:

• The endpoint requires auth credentials the agent cannot obtain (ask the user — allow skip)

If a request to either system fails to connect, ask the user to start the app before retrying. Do not silently skip — the user may have simply forgotten to start it.

Always ask the user before making any HTTP call that could have side effects.

Layer 3: Verification-Before-Completion Gate

Before marking ANY entity as migrated, the agent MUST have fresh evidence from the current session:

• Test command output showing pass count and exit code, OR
• HTTP comparison results showing a match, OR
• Explicit user approval to skip validation

Rules:

• No "should work", "looks correct", or "seems fine" — only evidence-backed claims
• The agent must state exactly what it verified and what the output was
• If evidence is insufficient, mark the entity as manual validation needed, not migrated
• Stale evidence from a previous session does not count — re-run validation if resuming

d. Update the Detail File

Update the entity's status in the unit's detail file (migration-plan/<unit-name>.md) and record validation evidence in that file's Validation Log table.

e. Update the Summary

When all entities in a unit are complete, update the unit's status in migration-plan.md to migrated. If some entities are pending, set the unit status to in progress.

5. Continue or Pause

After completing a unit, ask "What would you like to migrate next?" and suggest the next unit based on dependency order.

6. Batching

The default is one unit at a time. If the user says "keep going", "do them all", or similar, batch multiple units but still validate each entity individually before marking it as migrated.

Phase 4 — Complete

When all units in migration-plan.md are migrated, skipped, or manual validation needed:

1. Present a final summary from migration-plan.md:

- Total units migrated - Units marked as manual validation needed — read those specific detail files and list the entities that need attention with reasons - Units skipped (list them with reasons)

1. Suggest running the full test suite one final time to catch any integration issues
2. Note any manual validation items that still need human attention
3. If the source system had frontend code, suggest using the encore-frontend skill to reconnect the frontend to the new Encore backend (generate a typed API client, configure CORS, update base URLs)
4. Suggest removing migration-plan.md and migration-plan/ from the project once the user is satisfied with the migration

Asking Questions

Ask the user before acting when:

• Service boundaries are unclear — e.g., "These route files could be 1 service or 3 — how would you like to split them?"
• No clean Encore equivalent exists — e.g., Redis caching layer, custom middleware chains, WebSocket handlers
• Multiple valid migration strategies exist — present the options with tradeoffs
• Before making any HTTP call that could have side effects — always ask first
• Source code is ambiguous — when the agent is not confident about what the code does, ask rather than guess
• Source system appears to have changed — if files referenced in a detail file no longer exist or have changed significantly

Source System Protection

The source system must never be modified during migration. Follow these rules:

• Never modify source files — read them, don't edit them
• Never delete source files — even after migration is complete
• Never write to the source directory — all output goes to the Encore project
• Never run destructive commands against the source system (drop tables, delete queues, etc.)
• Ask before any HTTP call that mutates state on the source system (POST, PUT, DELETE)

If the user asks to "clean up" or "remove" the old system, confirm explicitly before taking any action. The source system may still be serving production traffic.

Edge Cases

Moving Endpoints Between Units

If the user realizes an endpoint belongs in a different migration unit:

1. Remove the endpoint row from the source unit's detail file
2. Add it to the target unit's detail file
3. Update endpoint counts in migration-plan.md

Splitting a Unit Mid-Migration

If a unit turns out to be too large while working on it:

1. Create a new detail file for the split-off portion (migration-plan/<new-unit>.md)
2. Move pending entities to the new file (already-migrated entities stay in the original)
3. Add the new unit to the migration-plan.md summary table
4. Insert it in the dependency order (same tier, after the original)

Monolith to Multiple Encore Services

A single migration unit might map to multiple Encore services. The detail file tracks the source grouping, but the "Notes" column can indicate the target Encore service. Ask during migration if the unit maps to one service or should be split across Encore services.

Source Code Changed Since Discovery

If files referenced in a detail file no longer exist or have changed significantly since discovery:

1. Flag the discrepancy to the user
2. Ask whether to update the detail file with the new state or skip the affected entities
3. If updating, re-assess complexity and adjust the plan accordingly

Troubleshooting

Common issues during migration and how to resolve them:

migration-plan.md Format

Use this exact template for the summary plan file. Fill in values from the discovery phase.

# Migration Plan

## Source System
- **Path:** <source system path>
- **URL:** <source system local URL>
- **Framework:** <detected framework>
- **Language:** <detected language>

## Frontend (Out of Scope)
- **Detected:** <Yes/No>
- **Directories:** <list of frontend directories, or "None">
- **Framework:** <frontend framework if detected, or "N/A">
- **Note:** <any framework server-side code that WAS included in migration units>

## Target System
- **Path:** <encore project path>
- **URL:** <encore local URL>
- **Type:** Encore.ts | Encore Go

## Migration Units

| Unit | Endpoints | DB Tables | Other | Complexity | Status |
|------|-----------|-----------|-------|------------|--------|

## Dependency Order
1. <ordered list of migration units>

Status values: pending, in progress, migrated, skipped, manual validation needed

Complexity values: Low (direct equivalent), Medium (requires restructuring), High (needs redesign)

Detail File Format

Create one file per migration unit at migration-plan/<unit-name>.md. Use this exact template:

# Migration Unit: <unit-name>

## Source
- **Files:** <list of source files in this unit>
- **Depends on:** <other migration units, with their current status>

## Endpoints
| Endpoint | Method | Path | Status | Notes |
|----------|--------|------|--------|-------|

## Database
| Table | Complexity | Status | Notes |
|-------|------------|--------|-------|

## Tests
- **Source tests:** <test files and count>
- **Migrated:** <count of migrated tests>

## Validation Log
| Entity | Tests | HTTP Match | Evidence | Status |
|--------|-------|------------|----------|--------|

Not all sections are required — omit sections that don't apply to a given unit (e.g., a secrets unit won't have Endpoints or Database sections).