Phoenix Code Review

Reviews Phoenix code for controller patterns, context boundaries, routing, and plugs. Use when reviewing Phoenix apps, checking controllers, routers, or cont...

Security Scan

The skill's files and runtime instructions match its stated purpose (Phoenix code review); it is instruction-only, requires no installs or credentials, and has no obvious mismatches.

Purpose & Capability

Name, description, and all included reference docs (controllers, contexts, plugs, routing) align with a Phoenix code-review helper. The skill requests no environment variables, binaries, or config paths that would be unrelated to code review.

Instruction Scope

The SKILL.md explicitly instructs the agent to open project source files and cite file:line for findings — this is expected for a code-review skill, but it grants the agent broad read access to the repository. The gates reference an external verification skill at ../review-verification-protocol/SKILL.md which is not included here; that dependency may block verification or produce vague behavior if missing. Overall scope is coherent with the purpose, but you should be aware it will inspect repository files and expects another review skill to be present.

Install Mechanism

No install spec and no code files: instruction-only (no code is written to disk or downloaded). This is the lowest-risk install posture.

Credentials

The skill declares no required environment variables, credentials, or config paths. Nothing requests unrelated secrets or system access.

Persistence & Privilege

always:false and default agent invocation are set. The skill does not request permanent presence or system-wide configuration changes; autonomy is allowed by default but is not combined with other elevated privileges here.

Assessment

This skill appears coherent and low-risk: it is a documentation-driven code-review helper that will read your project files to produce findings. Before using it, ensure you only grant the agent access to the repository you want reviewed, confirm whether the referenced ../review-verification-protocol skill exists (or provide an equivalent verification process), and be mindful that reported findings will include file paths and line numbers (which may expose code snippets). If you need stricter control, run the review in a sandboxed environment or limit the agent's filesystem access. Lastly, remember autonomous invocation is allowed by default — disable it if you prefer to run reviews manually.

Quick Reference

Issue Type	Reference
Bounded contexts, Ecto integration	references/contexts.md
Actions, params, error handling	references/controllers.md
Pipelines, scopes, verified routes	references/routing.md
Custom plugs, authentication	references/plugs.md

Review Checklist

Controllers

• Business logic in contexts, not controllers
• Controllers return proper HTTP status codes
• Action clauses handle all expected patterns
• Fallback controllers handle errors consistently

Contexts

• Contexts are bounded by domain, not technical layer
• Public functions have clear, domain-focused names
• Changesets validate all user input
• No Ecto queries in controllers

Routing

• Verified routes (~p sigil) used, not string paths
• Pipelines group related plugs
• Resources use only needed actions
• Scopes group related routes

Plugs

• Authentication/authorization via plugs
• Plugs are composable and single-purpose
• Halt called after sending response in plugs

JSON APIs

• Proper content negotiation
• Consistent error response format
• Pagination for list endpoints

Valid Patterns (Do NOT Flag)

• Controller calling multiple contexts - Valid for orchestration
• Inline Ecto query in context - Context owns its data access
• Using action_fallback - Centralized error handling pattern
• Multiple pipelines per route - Composition is intentional
• Plug.Conn.halt/1 without send - May be handled by fallback

Context-Sensitive Rules

Issue	Flag ONLY IF
Missing changeset validation	Field accepts user input AND no validation exists
Controller too large	More than 7 actions OR actions > 20 lines
Missing authorization	Route is not public AND no auth plug in pipeline

Gates (run in order; each step has a pass condition)

1. Anchored evidence — For every planned finding, open the source and note file path + line number from that read (not from memory or diff snippets alone). Pass: each finding cites path:line that you opened.
2. “Handled elsewhere” sweep — Before reporting “missing validation,” “missing auth,” or “wrong status,” search the router (pipelines/scopes), controller (action_fallback, plug), and relevant context for existing checks. Pass: you recorded whether handling exists elsewhere (yes + where, or no after search).
3. Verification protocol — Load and apply review-verification-protocol for the issue type. Pass: that skill’s pre-report checks for that finding class are satisfied before you write the finding.
4. Finding shape — Emit each issue as [FILE:LINE] ISSUE_TITLE with a one-line rationale tied to the cited code. Pass: every line matches that pattern.

Before Submitting Findings

Do not report until Gates above pass. For full anti-false-positive steps, follow review-verification-protocol.