review-changes

name	review-changes
description	Review all changes on the current branch. Use when the user asks for a code review or PR review.
disable-model-invocation	false
inputs	[{"id":"branch","type":"currentBranch","description":"The branch to review"}]

review-changes

Review all changes on the current branch compared to the upstream branch.

Context

{{branch}}

Determining the Base Branch

IMPORTANT: This repository uses develop as the integration branch, not master or main.

1. Use develop as the default base branch for all comparisons:

   git diff origin/develop...HEAD
   

2. Verify by checking GitHub workflows if unsure - look at .github/workflows/*.yml for pull_request.branches targets

3. Use the correct git diff syntax:

   • origin/develop...HEAD - shows changes in current branch since it diverged from develop
   • Run git log --oneline origin/develop..HEAD to see commits being reviewed

Pre-flight: Detect Change Types

Before launching subagents, analyze what files changed:

git diff origin/develop...HEAD --name-only

Conservative skip rules (only skip when 100% safe):

Reviewer	Run	Notes
reviewer-prettier	✅ Always	Formats .md, .json, .ts, .tsx, etc.
reviewer-versioning	✅ Always	Fast validation, can fail early
reviewer-changelog	✅ Always	Fast validation, can fail early
reviewer-dependencies	Conditional	Run if ANY package.json changed
reviewer-typescript	Conditional	Skip ONLY if NO .ts/.tsx in packages/ changed
reviewer-eslint	Conditional	Skip ONLY if NO .ts/.tsx in packages/ changed
reviewer-tests	Conditional	Skip ONLY if NO .ts/.tsx in packages/ changed
reviewer-complexity	Conditional	Skip ONLY if NO .ts/.tsx in packages/ changed

When in doubt, run the check. Fast failures are better than missed issues.

Execution Strategy

IMPORTANT: Launch ALL applicable subagents in a SINGLE parallel batch. Do NOT wait for one group to finish before starting another.

In one tool call batch, launch all applicable reviewers:

• reviewer-prettier (always)
• reviewer-versioning (always)
• reviewer-changelog (always)
• reviewer-dependencies (if any package.json changed)
• reviewer-typescript (if .ts/.tsx in packages/ changed)
• reviewer-eslint (if .ts/.tsx in packages/ changed)
• reviewer-tests (if .ts/.tsx in packages/ changed)
• reviewer-complexity (if .ts/.tsx in packages/ changed)

Note: reviewer-dependencies checks changelog documentation but does NOT create/modify changelogs. If both changelog and dependency changes exist, both reviewers run in parallel - the dependency reviewer only reads existing changelogs.

Analysis Required

Check for:

Code Quality & Bugs:

• Potential bugs, edge cases, or runtime errors
• Code smells, anti-patterns, or violations of repository rules (check .cursor/rules/*)
• Newly added TODO/FIXME comments that should be addressed
• Suspicious or unclear code changes
• Business logic changes that need scrutiny

Standards & Compliance:

• Delegate to reviewer-typescript subagent to check for TypeScript errors
• Delegate to reviewer-eslint subagent to check for linting violations
• Delegate to reviewer-prettier subagent to check for formatting issues
• Breaking changes to public API surface
• Package exports changes
• Type definition changes

FuryStack Library-Specific:

• Public API changes (exported functions, tokens, types)
• Breaking changes detection
• Dependency injection patterns (defineService, defineStore, defineDataSet, injector.get/bind/invalidate/createScope, useSystemIdentityContext, throw-by-default store tokens)
• Observable patterns (ObservableValue, subscriptions)
• Disposable resources (Symbol.dispose, Symbol.asyncDispose, factory onDispose)
• Package structure and organization

Shades Patterns:

• Flag useState() used only for CSS-representable states (hover, focus, active)

• Recommend using css property with pseudo-selectors instead:

  // ❌ Anti-pattern to flag
  const [isHovered, setIsHovered] = useState('hover', false)
  <div onMouseEnter={() => setIsHovered(true)} style={{ opacity: isHovered ? 1 : 0.7 }} />
  
  // ✅ Recommend instead
  css: { opacity: '0.7', '&:hover': { opacity: '1' } }
  

• Static style props in Shade definitions should use css instead

• Flag usage of element in render function destructuring or body -- it was removed, use useHostProps or useRef instead

• Flag usage of onAttach or onDetach in ShadeOptions -- they were removed, use useDisposable instead

• Flag direct DOM manipulation inside render() that should use useHostProps (e.g. this.setAttribute(...), this.style.xxx = ...)

• Recommend useRef + queueMicrotask for deferred child element access instead of direct DOM queries

Complexity Audit:

• Delegate to reviewer-complexity subagent to flag overgrown components, services, and REST actions introduced or worsened by the branch (heuristics from .cursor/rules/COMPLEXITY.mdc)

Testing & Coverage:

• Delegate to reviewer-tests subagent to run unit tests and assess coverage
• Missing test coverage for new/changed public APIs
• Test quality and edge case coverage

Performance & Security:

• Performance concerns (memory leaks, missing disposal)
• Security vulnerabilities or data exposure

Versioning & Changelog:

• Delegate to reviewer-versioning subagent to validate version bumps
• Delegate to reviewer-changelog subagent to validate changelog entries

Dependencies:

• Delegate to reviewer-dependencies subagent to validate dependency consistency across packages

Documentation:

• Missing or outdated documentation for public APIs
• README updates for new features
• Breaking change documentation

ESLint Rule Opportunities:

Look for patterns in the diff that suggest a new custom ESLint rule should be added to @furystack/eslint-plugin. Signals to watch for:

• Repeated anti-pattern fixes — the same kind of correction applied across multiple files (e.g., replacing X with Y in 5+ places). If a human has to remember to avoid it, a lint rule can enforce it.
• New conventions introduced — if the branch establishes a new "always do X instead of Y" pattern, it may be worth codifying as a rule.
• Manual enforcement via comments — code comments like "don't do X", "always use Y here", or "TODO: lint rule for this" are direct signals.
• Domain-specific patterns — especially around DI (defineService, defineStore, defineDataSet, token-based resolution), Observables (ObservableValue, subscriptions, disposal), Shades rendering (render hooks, JSX patterns), REST actions (RequestError, validation), and data access (DataSetToken, StoreToken).

When evaluating a potential rule:

1. Check existing rules first — read packages/eslint-plugin/src/rules/ to verify the pattern isn't already covered. The plugin currently has rules for: DI consistency, observable disposal, Shades render hooks, REST action validation, custom element naming, removed APIs, and more.
2. Determine the config — would it belong in recommended (general FuryStack patterns) or shades (Shades UI framework patterns)?
3. Assess feasibility — can the pattern be reliably detected via AST analysis? Patterns that depend on runtime behavior or external context are poor candidates.
4. Consider false positives — a good rule has a low false-positive rate. If the pattern has many legitimate exceptions, it may not be worth a rule.

Subagent Output Rules

IMPORTANT: Only include subagent results in the final output if they found errors or issues.

• If reviewer-typescript passes → Do NOT mention it in the output
• If reviewer-eslint passes → Do NOT mention it in the output
• If reviewer-prettier passes → Do NOT mention it in the output
• If reviewer-tests passes → Do NOT mention it in the output
• If reviewer-versioning passes → Do NOT mention it in the output
• If reviewer-changelog passes → Do NOT mention it in the output
• If reviewer-dependencies passes → Do NOT mention it in the output
• If reviewer-complexity passes → Do NOT mention it in the output

Only report subagent findings when they detect actual problems.

Output Format

1. Summary: Brief overview of changes (2-3 sentences max)

2. Issues by Priority:

• 💀 Critical: Must fix before merge
• 🔥 High: Should fix before merge
• 🤔 Medium: Consider addressing
• 💚 Low: Nice to have

For each issue, be specific: package, file, line, problem, suggested fix.

3. Test Coverage: Assess coverage quality. Warn if public APIs lack tests.

4. Breaking Changes: List any breaking changes and suggest migration guide if needed.

5. ESLint Rule Suggestions: If any patterns were identified that could become custom ESLint rules, list them here. For each suggestion include:

• Pattern: What the rule would enforce (with a brief before/after code example if helpful)
• Config: recommended or shades
• Rationale: Why this is worth automating (frequency of occurrence, risk of mistakes, etc.)

Omit this section entirely if no rule opportunities were found.

6. Changelog: Generate a short, consistent changelog as a copyable markdown code block.

Format:

• Use present tense, imperative mood (e.g., "Add", "Fix", "Update", "Remove")
• One line per logical change (max 5-7 lines)
• Group by type if multiple changes: Features, Fixes, Refactors
• No verbose descriptions - keep each line under 80 characters

Example:

- Add user profile validation
- Fix observable memory leak in cache
- Update dependency injection patterns

7. Pull Request Description: Generate as a copyable markdown code block with:

• Relevant emoji per header
• Brief description of what the PR does
• Remaining Tasks checklist generated from found issues (if any), grouped by priority:

## 📋 Remaining Tasks

### 💀 Critical

- [ ] Fix TypeScript error in `packages/core/src/foo.ts:42`

### 🔥 High

- [ ] Add missing test for `handleSubmit` function

### 🤔 Medium

- [ ] Address memory leak in observable subscription

Omit empty priority sections. If no issues found, omit the entire Remaining Tasks section.

Style: Be critical, specific, and concise. Focus on library quality and API surface area. If unsure, ask for clarification.