Dangi Doctor 🩺 #

Your Flutter app's personal physician.

Dangi Doctor connects to your live Flutter app, crawls every screen automatically, finds real bugs through static analysis, measures performance, and generates Flutter integration tests — all without you writing a single line of test code.

What's new in v0.3.0 #

The weekly knowledge updater works again — all the Flutter doc sources had moved and were silently 404ing; they're repaired, the updater now fails loudly instead of shipping placeholders, and each release bundles current knowledge.
Generated tests compile — the interaction and performance tests referenced a helper that was never emitted; that's fixed, along with import paths, dynamic keys, and Cupertino-app support.
Crawler correctness (verified on a live Android device) — accurate screen naming through pushed route stacks, navigation that returns home without exiting the app, whole-word dialog-button matching, and coordinate re-resolution after returning from a child screen. It also no longer kills unrelated flutter run/adb sessions on your machine.
Real performance numbers — frames read from Flutter.Frame events, actual memory usage, and a jank budget that follows the device's real refresh rate (120 Hz → 8.3 ms, not a hardcoded 16 ms).
Faster, safer AI diagnosis — cross-screen prompt caching (~90% cheaper input), real retries/timeouts, current model defaults, and the Gemini key no longer travelling in the request URL.
Proper CLI — --help, --version, --project, --vm-url, --device, --no-ai; honest exit codes; safe to run in CI.

What it does #

1. Live app crawling #

Connects to your running Flutter app via the Dart VM service. Walks every reachable screen by tapping navigation triggers (bottom nav, buttons, drawers). No emulator required — works on real physical devices.

2. Static analysis — detects real bugs before they reach production #

Scans your lib/ source code for patterns that cause crashes at runtime:

Bug type	What it catches
`late_field_double_init`	`late` field assigned in a method called from `didChangeDependencies()` without a guard → `LateInitializationError`
`setState_after_dispose`	`setState()` in async method without `if (mounted)` → "setState called after dispose"
`stream_subscription_leak`	`StreamSubscription` field with no `.cancel()` in `dispose()` → memory leak
`build_side_effects`	`setState()` or `await` directly inside `build()` → infinite rebuild loop

3. Generated integration tests — with exact fix instructions #

For every screen and every detected bug, Dangi Doctor writes Flutter integration tests directly into your project at integration_test/dangi_doctor/. Each failing test tells you:

Exact file and line number
Plain-English explanation of the bug
Copy-pasteable fix code

BUG: pages/SplashScreen/splashScreenPage.dart:74 —
late field `_appLinks` double-init in `initDeepLinks()`

━━━ BUG DETECTED ━━━
File: pages/SplashScreen/splashScreenPage.dart:74

Problem:
Late field `_appLinks` is assigned in `initDeepLinks()` which is called
from `didChangeDependencies()` without an initialization guard...

Fix:
  bool _initDeepLinksCalled = false;

  @override
  void didChangeDependencies() {
    super.didChangeDependencies();
    if (!_initDeepLinksCalled) {
      _initDeepLinksCalled = true;
      initDeepLinks();
    }
  }
━━━━━━━━━━━━━━━━━━━

4. HTML health report #

After every diagnosis, Dangi Doctor opens a health report in your browser automatically.

Overall health score (0–100)
Per-screen performance grade (A–F) with build time, jank %, memory
All issues listed with severity (error / warning / info)
Static analysis bugs table with file locations and fix code
Saved to .dangi_doctor/report_<timestamp>.html

5. AI diagnosis (optional) #

If you have a Claude, OpenAI, Gemini, Groq, or Ollama API key, Dangi Doctor gives each screen a written diagnosis — what's wrong, why it matters, and a prioritised fix list.

The AI is powered by a 3-layer knowledge system that stays current automatically:

Layer 1 — Flutter official knowledge: widget lifecycle, performance rules, state management error patterns for Provider / BLoC / Riverpod / GetX, common Flutter errors by category, and the last 5 Flutter version changelogs. Updated weekly by pulling directly from the Flutter GitHub repo.
Layer 2 — Community anti-patterns: most common mistakes found across thousands of real production Flutter apps.
Layer 3 — Your project: auto-detected on first run. State management, dependencies, folder structure, naming conventions, largest files. Stored at .dangi_doctor/project.json.

Requirements #

Dart SDK ≥ 3.0
Flutter project with a physical Android device or emulator
adb installed (Android Debug Bridge)

Installation #

dart pub global activate dangi_doctor

Or add to your Flutter project's dev dependencies:

dev_dependencies:
  dangi_doctor: ^0.3.0

Usage #

From the root of your Flutter project:

cd /path/to/your/flutter/app
dangi_doctor

Dangi Doctor auto-detects your project from the current directory. If you need to point it at a different path, use the DANGI_PROJECT env var.

On first run, Dangi Doctor will ask how to connect:

┌─────────────────────────────────────────────┐
│  How do you want to connect?                │
│                                             │
│  1. Launch app now (Dangi Doctor runs it)   │
│  2. App already running — paste VM URL      │
└─────────────────────────────────────────────┘

Choose 1 to let Dangi Doctor launch your app, or 2 to paste the VM service URL from a running flutter run --debug session.

Command-line options #

Flag	Description
`--project <path>`	Flutter project path (overrides `DANGI_PROJECT` and cwd auto-detection)
`--vm-url <ws://…>`	VM service URL of an already-running app (skips the connect prompt)
`--device <id>`	ADB device id for widget taps (auto-detected if omitted)
`--no-ai`	Crawler + static analysis only; skip AI diagnosis
`--version`	Print the version and exit
`--help`	Print usage and exit

Fully non-interactive runs are supported for CI — pass --project, --vm-url, and --no-ai and Dangi Doctor never prompts; a crash exits non-zero.

Run generated tests #

flutter test integration_test/dangi_doctor/<screen>_smoke_test.dart \
  -d <device_id>

With auth token (for apps requiring login):

flutter test integration_test/dangi_doctor/<screen>_smoke_test.dart \
  --dart-define=TEST_TOKEN=your_token \
  -d <device_id>

AI diagnosis — free options #

You do not need a paid subscription to get AI diagnosis.

Provider	Cost	How to get a key
Groq	Free tier (14,400 req/day)	Sign up at `console.groq.com` — no credit card
Ollama	Free, runs locally	Install from `ollama.com`, run `ollama pull llama3.1`
Claude	Paid	`console.anthropic.com`
OpenAI	Paid	`platform.openai.com`
Gemini	Paid	`aistudio.google.com`

If no API key is set, Dangi Doctor asks at runtime which provider to use. Choose Groq for the quickest free setup, or Ollama for fully offline diagnosis.

Environment variables #

Variable	Description
`DANGI_PROJECT`	Path to your Flutter project (auto-detected from cwd; `--project` takes precedence)
`ANTHROPIC_API_KEY`	Claude API key for AI diagnosis (`CLAUDE_API_KEY` also accepted)
`OPENAI_API_KEY`	OpenAI API key for AI diagnosis
`GEMINI_API_KEY`	Gemini API key for AI diagnosis
`GROQ_API_KEY`	Groq API key for AI diagnosis

Model choices default to current models but are all overridable, so a model rename never requires an update:

Variable	Default
`DANGI_CLAUDE_MODEL`	`claude-opus-4-8`
`DANGI_OPENAI_MODEL`	`gpt-4o`
`DANGI_GEMINI_MODEL`	`gemini-2.5-pro`
`DANGI_GROQ_MODEL`	`llama-3.1-8b-instant`
`DANGI_OLLAMA_MODEL`	`llama3.1`
`DANGI_OLLAMA_URL`	`http://localhost:11434`
`DANGI_MAX_TOKENS`	`4096`

How it works #

dart run dangi_doctor
        │
        ├── 1. Detect AI provider (Claude / OpenAI / Gemini / Groq / Ollama)
        │
        ├── 2. Connect to Flutter app via Dart VM service WebSocket
        │         └── Port remapping: reads /proc/net/tcp on device to find
        │             the real VM port, then adb forward tcp:8181 tcp:<actual>
        │
        ├── 3. Wait for splash screen to dismiss
        │
        ├── 4. Static analysis
        │         └── Scan lib/ for KnownRisk patterns + detect routes
        │         └── LateInitializationError, setState after dispose,
        │             stream leaks, build() side effects
        │
        ├── 5. Crawl all screens
        │         └── Tap nav triggers (bottom nav, buttons, drawers)
        │         └── Capture widget tree per screen via VM service
        │         └── Measure frame performance via Flutter.Frame events
        │         └── Detect widget tree issues (nesting, anti-patterns)
        │
        ├── 6. AI diagnosis per screen (if API key present)
        │         └── 3-layer knowledge prompt → Claude/GPT/Gemini
        │
        ├── 7. Generate integration tests
        │         └── integration_test/dangi_doctor/<screen>_smoke_test.dart
        │         └── integration_test/dangi_doctor/<screen>_interaction_test.dart
        │         └── integration_test/dangi_doctor/<screen>_perf_test.dart
        │         └── integration_test/dangi_doctor/test_helper.dart
        │
        └── 8. Generate HTML report → open in browser
                  └── .dangi_doctor/report_<timestamp>.html

Knowledge auto-update #

Layer 1 knowledge is bundled as a Dart constant in the package. A weekly GitHub Actions workflow refreshes it in the repo from the Flutter GitHub repo and flutter.dev docs, and every dangi_doctor release ships the knowledge current at release time — so dart pub upgrade dangi_doctor picks up whatever the newest published version bundles. Layer 2 (community anti-patterns) is maintained in knowledge/community_patterns.md — PRs welcome.

To update manually (for contributors):

dart run tool/update_knowledge.dart

Output files #

After a diagnosis run, your Flutter project will contain:

integration_test/
  dangi_doctor/
    test_helper.dart                  ← shared setup: Firebase init, auth injection
    <screen>_smoke_test.dart          ← launch test + bug-specific targeted tests
    <screen>_interaction_test.dart    ← tap every button, verify navigation
    <screen>_perf_test.dart           ← frame timing assertions

.dangi_doctor/
  project.json                        ← Layer 3 project fingerprint
  report_<timestamp>.html             ← HTML health report
  vm_url.txt                          ← cached VM URL for reconnection

License #

MIT

dangi_doctor 0.3.1
dangi_doctor: ^0.3.1 copied to clipboard

Metadata

Dangi Doctor 🩺 #

What's new in v0.3.0 #

What it does #

1. Live app crawling #

2. Static analysis — detects real bugs before they reach production #

3. Generated integration tests — with exact fix instructions #

4. HTML health report #

5. AI diagnosis (optional) #

Requirements #

Installation #

Usage #

Command-line options #

Run generated tests #

AI diagnosis — free options #

Environment variables #

How it works #

Knowledge auto-update #

Output files #

License #

← Metadata

Documentation

Publisher

Weekly Downloads

Metadata

Topics

License

Dependencies

More

dangi_doctor 0.3.1 dangi_doctor: ^0.3.1 copied to clipboard

Metadata

Dangi Doctor 🩺 #

What's new in v0.3.0 #

What it does #

1. Live app crawling #

2. Static analysis — detects real bugs before they reach production #

3. Generated integration tests — with exact fix instructions #

4. HTML health report #

5. AI diagnosis (optional) #

Requirements #

Installation #

Usage #

Command-line options #

Run generated tests #

AI diagnosis — free options #

Environment variables #

How it works #

Knowledge auto-update #

Output files #

License #

← Metadata

Documentation

Publisher

Weekly Downloads

Metadata

Topics

License

Dependencies

More

dangi_doctor 0.3.1
dangi_doctor: ^0.3.1 copied to clipboard