Rebuild Solid skill + add AGENTS.md template for AI assistants

.gitignore

@@ -48,3 +48,6 @@ lcov.info
 .firebase
 .fvmrc
 .supermaven
+
+# skill-creator run_loop / eval workspaces (ephemeral artifacts)
+solid-workspace/

analysis_options.yaml

@@ -1,5 +1,10 @@
 include: package:very_good_analysis/analysis_options.yaml
 analyzer:
+  exclude:
+    # Solid skill eval fixtures: template project skeletons used by
+    # skill-creator's eval runner. They reference flutter_lints, which the
+    # workspace doesn't resolve, and they aren't real source to be analyzed.
+    - skills/solid/evals/files/**
   errors:
     always_put_required_named_parameters_first: ignore
     package_names: ignore

docs/src/content/docs/guides/getting-started.mdx

@@ -70,11 +70,39 @@ import { Steps } from '@astrojs/starlight/components';
    ```
 </Steps>
 
-<Aside title="AI assistant setup (optional)">
-If you use Claude Code, Cursor, Codex, GitHub Copilot, or another supported agent, install the Solid skill so your assistant follows Solid's conventions (edit `source/`, never `lib/`; correct annotation targets; setup steps): `npx skills add nank1ro/solid`. Or copy `skills/solid/SKILL.md` from the [Solid repo](https://github.com/nank1ro/solid) into your editor's skill location.
+## AI assistant setup
 
-If your tool fetches documentation over HTTP (Cursor `@docs`, ChatGPT custom GPTs, claude.ai web search, …), point it at [`/llms-full.txt`](https://solid.mariuti.com/llms-full.txt) — the full docs as a single LLM-friendly file. A short index lives at [`/llms.txt`](https://solid.mariuti.com/llms.txt).
-</Aside>
+If you're using an AI coding assistant (Claude Code, Cursor, Codex, GitHub Copilot, Amp, OpenHands, etc.), these three steps teach it the inverted `source/`-vs-`lib/` rule and Solid's annotation contract — so it stops trying to edit `lib/` and silently losing your work.
+
+<Steps>
+
+1. Drop `AGENTS.md` at your app root.
+
+   Most AI coding tools auto-load `AGENTS.md` at session start, so the rule applies to every interaction — even short or routine-looking ones — without depending on skill triggering.
+
+   ```bash
+   curl -o AGENTS.md https://raw.githubusercontent.com/nank1ro/solid/main/skills/solid/assets/AGENTS.md
+   ```
+
+   If your tool only looks for `CLAUDE.md`, symlink it:
+
+   ```bash
+   ln -s AGENTS.md CLAUDE.md
+   ```
+
+2. Install the Solid skill.
+
+   ```bash
+   npx skills add nank1ro/solid
+   ```
+
+   Or copy `skills/solid/` from the [Solid repo](https://github.com/nank1ro/solid) into your editor's skill location. The skill gives the agent deeper guidance (full annotation contract, scripts, troubleshooting) when it triggers on Solid-related work. `AGENTS.md` is the always-loaded baseline; the skill is the on-demand reference.
+
+3. Point HTTP-docs tools at the LLM-friendly bundle (optional).
+
+   If your tool fetches documentation over HTTP (Cursor `@docs`, ChatGPT custom GPTs, claude.ai web search, …), point it at [`/llms-full.txt`](https://solid.mariuti.com/llms-full.txt) — the full docs as a single LLM-friendly file. A short index lives at [`/llms.txt`](https://solid.mariuti.com/llms.txt).
+
+</Steps>
 
 ## How it works
 
@@ -136,6 +164,25 @@ dart run build_runner watch --delete-conflicting-outputs
 - Press `r` in the `flutter run` terminal after `build_runner` emits.
 - Use [`dashmonx`](https://pub.dev/packages/dashmonx), which wraps `flutter run` and triggers hot reload automatically when files under `lib/` change. Any `flutter run` flag passes through, e.g. `dashmonx -d chrome` for a web target.
 
+### Clean up after generation
+
+Solid prioritises producing correct, runnable code over polishing it. The generated `lib/` output may miss some `const` opportunities, leave unused imports, or pick a non-preferred import form. Run `dart fix` after `build_runner` to apply the lint-driven fixes (`prefer_const_constructors`, `unnecessary_import`, `prefer_relative_imports`, …):
+
+```bash
+dart fix --apply
+```
+
+In CI, chain the two commands so the generated output is always lint-clean:
+
+```bash
+dart run build_runner build --delete-conflicting-outputs
+dart fix --apply
+```
+
+<Aside>
+The Solid skill (installed via the AI assistant setup above) ships a `verify.sh` helper that runs both in one shot and reports PASS/FAIL.
+</Aside>
+
 <Aside>
 The next chapters assume you have set up Solid in your Flutter project, created the `source` folder, and are running `dart run build_runner watch --delete-conflicting-outputs` to regenerate `lib/` on every change.
 </Aside>

example/analysis_options.yaml

@@ -6,6 +6,7 @@ analyzer:
     invalid_annotation_target: ignore
     avoid_print: ignore
     unnecessary_statements: ignore
+    always_use_package_imports: ignore
 linter:
   rules:
     public_member_api_docs: false

example/lib/main.dart

@@ -1,7 +1,8 @@
-import 'package:example/counter.dart';
 import 'package:flutter/material.dart';
 import 'package:flutter_solidart/flutter_solidart.dart';
 
+import 'counter.dart';
+
 void main() {
   SolidartConfig.autoDispose = false;
   runApp(const MaterialApp(home: CounterPage()));

example/source/main.dart

@@ -1,8 +1,9 @@
-import 'package:example/counter.dart';
 import 'package:flutter/material.dart';
 import 'package:flutter_solidart/flutter_solidart.dart';
 
+import 'counter.dart';
+
 void main() {
   SolidartConfig.autoDispose = false;
-  runApp(const MaterialApp(home: CounterPage()));
+  runApp(MaterialApp(home: CounterPage()));
 }