georgeguimaraes/elixir-agent-tools/phoenix
Claude Code and agent skill by georgeguimaraes/elixir-agent-tools
- community
Checking that the SKILL.md file is still in the repository.
Build and debug Phoenix web interfaces and HTTP endpoints. Use for LiveView lifecycle and data loading, components, forms, routes, controllers, Plug, channels, and PubSub. Use ecto for changesets, queries, and persistence behind those interfaces.
Install
npx vskill@latest install georgeguimaraes/elixir-agent-tools/phoenixSKILL.md
Phoenix
Structure Phoenix interfaces, load LiveView data, and scope real-time updates.
Where to Load Data: mount vs handle_params
Default: load data in mount/3.
Yes, mount runs twice on initial load (HTTP dead render + WebSocket connect). So does handle_params/3. That's the LiveView lifecycle, not a bug to route around. Moving queries from mount to handle_params does not dedupe them.
Use handle_params/3 for data that changes on live navigation (push_patch / <.link patch={...}>). mount does not re-run on patches, handle_params does.
When the initial double-load actually matters, the real tools are:
connected?(socket)to gate work to the connected render (loses SEO / no-JS rendering)assign_async/3to load after mount returns, in a separate processassign_new/3to reuse values already set onconn.assignsby upstream Plugs (e.g.:current_user), or shared from a parent LiveView. It does not dedupe arbitrary work across the dead/connected boundary: the function still runs on connected mount.
Scopes: Security-First Pattern (Phoenix 1.8+)
Scopes address OWASP #1 vulnerability: Broken Access Control. Authorization context is threaded automatically—no more forgetting to scope queries.
PubSub Topics Must Be Scoped
Unscoped topics = data leaks between tenants.
External Polling: GenServer, Not LiveView
Bad: Every connected user makes API calls (multiplied by users). Good: Single GenServer polls, broadcasts to all via PubSub.
Components Receive Data, LiveViews Own Data
- Functional components: Display-only, no internal state
- LiveComponents: Own state, handle own events
- LiveViews: Full page, owns URL, top-level state
Async Data Loading
Use assign_async/3 for data that can load after mount:
Gotchas from Core Team
LiveView terminate/2 Requires trap_exit
terminate/2 only fires if you're trapping exits—which you shouldn't do in LiveView.
Fix: Use a separate GenServer that monitors the LiveView process via Process.monitor/1, then handle :DOWN messages to run cleanup.
start_async Duplicate Names: Later Wins
Calling start_async with the same name while a task is in-flight: the later one wins, the previous task's result is ignored.
Fix: Call cancel_async/3 first if you want to abort the previous task.
Channel Intercept Socket State is Stale
The socket in handle_out intercept is a snapshot from subscription time, not current state.
Why: Socket is copied into fastlane lookup at subscription time for performance.
Fix: Use separate topics per role, or fetch current state explicitly.
CSS Class Precedence is Stylesheet Order
When merging classes on components, precedence is determined by stylesheet order, not HTML order. If btn-primary appears later in the compiled CSS than bg-red-500, it wins regardless of HTML order.
Fix: Use variant props instead of class merging.
Upload Content-Type Can't Be Trusted
The :content_type in %Plug.Upload{} is user-provided. Always validate actual file contents (magic bytes) and rewrite filename/extension.
Read Body Before Plug.Parsers for Webhooks
To verify webhook signatures, you need the raw body. But Plug.Parsers consumes it.
Don't use preserve_req_body: true—it keeps the entire body in memory for ALL requests.
Red Flags - STOP and Reconsider
- Loading patch-mutable data in mount/3 instead of handle_params/3
- Unscoped PubSub topics in multi-tenant app
- LiveView polling external APIs directly
- Using terminate/2 for cleanup (won't fire without trap_exit)
- Calling start_async with same name without cancel_async first
- Relying on socket.assigns in Channel intercepts (stale!)
- CSS class merging for component customization (use variants)
- Trusting
%Plug.Upload{}.content_typefor security
Any of these? Re-read the Gotchas section.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingplugins/elixir-dev/skills/phoenix/SKILL.md
Popularity
- Repository stars
- 185
- Repository forks
- 15
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/georgeguimaraes/elixir-agent-tools/phoenix)