PIU Backend Sync

Analyzes a backend repository, discovers API routes, and creates PIU entities (project, collections, requests, environments) via MCP. Tracks git commit SHA for incremental re-syncs.

CLI Scripts

This skill bundles scripts/piu.ts and scripts/detect.ts (relative to this SKILL.md):

bun scripts/piu.ts <command> [args...]     # MCP client (46 tools)
bun scripts/detect.ts /path/to/repo        # Framework detection → JSON

For full tool reference, see the piu-mcp skill.

Step 0: Re-Sync Detection

For previously imported projects, check if the repo has changed:

# Check if project exists
bun scripts/piu.ts list-projects

# Compare commits
bun scripts/piu.ts diff-sync PROJECT_ID /path/to/repo

For incremental sync, get the current PIU state and compare:

bun scripts/piu.ts tree PROJECT_ID
# Then re-extract changed routes and create/update/flag as needed
bun scripts/piu.ts update-project '{"project_id":"...","source_commit_id":"NEW_COMMIT"}'

Step 1: Clone & Detect Framework

# Remote repo
TMPDIR=$(mktemp -d /tmp/piu-sync-XXXXX)
git clone --depth 1 <url> "$TMPDIR/repo"
REPO="$TMPDIR/repo"

# Or local repo
REPO=/path/to/repo

COMMIT=$(git -C "$REPO" rev-parse HEAD)
DETECT=$(bun scripts/detect.ts "$REPO")
# Returns: {"framework":"express","port":3000,"router_files":["routes/api.js"]}

Supported frameworks: Hyperf, Laravel, Express, Fastify, NestJS, Hono, FastAPI, Django, Flask, Gin, Echo, Fiber, Axum, Actix, Rails, Spring.

Step 2: Route Extraction

Read each router file detected in Step 1 and extract routes by framework pattern:

For each route: extract HTTP method, URL path, handler name, group/prefix.

Step 3: Create PIU Entities

3a. Project + Environment

bun scripts/piu.ts create-project '{"name":"PROJECT_NAME","description":"Imported from <url>","source_repo_url":"<url>","source_commit_id":"COMMIT","backend_type":"FRAMEWORK"}'
# → returns {"id": "PROJECT_ID", ...}

bun scripts/piu.ts create-env '{"project_id":"PROJECT_ID","name":"Development","host":"http://localhost:PORT"}'

3b. Collections (batch)

cat <<'EOF' | bun scripts/piu.ts batch-collections
[
  {"project_id":"PROJECT_ID","name":"Users","path_prefix":"/users","description":"User management","source_commit_id":"COMMIT"},
  {"project_id":"PROJECT_ID","name":"Auth","path_prefix":"/auth","description":"Authentication","source_commit_id":"COMMIT"}
]
EOF

3c. Requests (batch)

For large imports, write one JSON file per collection:

cat /tmp/piu-routes/users.json | bun scripts/piu.ts batch-requests

Format: [{"collection_id":"...","name":"List Users","method":"GET","url":"/list","description":"..."}]

For 500+ routes, use parallel subagents each processing a subset.

3d. Environment Setup

bun scripts/piu.ts set-vars '{"environment_id":"ENV_ID","variables":[{"key":"token","value":"your-auth-token","enabled":true}]}'
bun scripts/piu.ts activate-env '{"environment_id":"ENV_ID","project_id":"PROJECT_ID"}'

Step 4: Model Extraction

After creating requests, extract data models from controller schemas.

Shared base models

Create reusable models first:

cat <<'EOF' | bun scripts/piu.ts batch-models
{
  "project_id": "PROJECT_ID",
  "models": [
    {"name":"PaginationParams","description":"Common pagination","fields":[{"name":"page","field_type":"integer","required":false,"example":"1"},{"name":"per_page","field_type":"integer","required":false,"example":"20"}]},
    {"name":"ApiResponse","description":"Standard wrapper","fields":[{"name":"code","field_type":"integer","required":true,"example":"0"},{"name":"message","field_type":"string","required":true,"example":"success"},{"name":"data","field_type":"object","required":false}]}
  ]
}
EOF

Per-endpoint models

For each collection, read controller source and extract request/response schemas:

Link models to requests

cat <<'EOF' | bun scripts/piu.ts batch-links
[
  {"request_id":"REQ_ID","model_type":"request","model_id":"MODEL_ID"},
  {"request_id":"REQ_ID","model_type":"response","model_id":"RESP_MODEL_ID"}
]
EOF

Step 5: API Documentation

Every request description should be a complete markdown document:

````markdown

POST /auth/login

Authenticate user credentials and return a JWT token.

Parameters

Request Body

{"username": "admin", "password": "secret123"}

Response

Returns {code: 0, data: {token: "...", expires_in: 3600}}.

Notes

• Rate limited to 5 attempts per minute
• See also: POST /auth/refresh

````

Use batch-update-bodies to apply descriptions:

cat updates.json | bun scripts/piu.ts batch-update-bodies

Step 6: Verification

# Project overview
bun scripts/piu.ts overview PROJECT_ID

# Full tree
bun scripts/piu.ts tree PROJECT_ID

# Execute all GET endpoints (requires running backend + active env)
bun scripts/piu.ts verify PROJECT_ID

# Model visualization
bun scripts/piu.ts model-mermaid PROJECT_ID

# Sync status
bun scripts/piu.ts sync-status PROJECT_ID

# Changelog audit
bun scripts/piu.ts changelog '{"entity_type":"project","entity_id":"PROJECT_ID","limit":20}'

# Search for specific endpoints
bun scripts/piu.ts search PROJECT_ID "/login" POST

# API surface summary
bun scripts/piu.ts api-surface PROJECT_ID

Step 7: Cleanup & Report

rm -rf "$TMPDIR"  # Only if cloned to temp dir

Print summary:

## Backend Sync Complete

**Repository:** <url>
**Commit:** <short_sha>
**Framework:** <detected>
**Project ID:** <id>

### Created:
- 1 project, 1 environment
- <N> collections, <M> requests
- <X> models linked to <Y> requests

### Methods: GET: N | POST: N | PUT: N | DELETE: N

Notes

• Always use search before creating to avoid duplicates on re-sync
• Set sourcecommitid on every entity for future re-sync
• One collection per router/controller/blueprint, not one flat list
• For monorepo structures, ask which service to import
• Prefer reading route definitions over OpenAPI/Swagger specs (those can be outdated)
• For re-syncs, run diff-sync first to avoid full re-imports
• Use api-surface for a quick summary of all endpoints
• Use find-related to explore entity relationships