feat(pr-review): Add missing documentation and changeset checks

.claude/agents/pr-review-docs.md

@@ -54,21 +54,31 @@ Review documentation files for compliance with **write-docs skill standards**.
 # Workflow
 
 1. **Review the PR context** — The diff, changed files, and PR metadata are available via your loaded `pr-context` skill
-2. **Read each file** using Read tool
-3. **Evaluate against write-docs skill** - use the skill's verification checklist as your rubric:
+2. **Check for missing documentation updates** (MANDATORY FIRST STEP):
+   - If PR modifies published packages or adds features BUT lacks documentation changes, create a HIGH confidence CRITICAL finding
+   - Trigger patterns that require documentation (per AGENTS.md):
+     - New UI components: `*.tsx` files in `agents-manage-ui/src/components/`
+     - New API routes: new files in `agents-api/src/domains/*/routes/`
+     - Schema changes: modifications to `*-schema.ts` files or files in `packages/agents-core/src/validation/`
+     - New package exports: changes to `package.json` `exports` field
+     - New SDK patterns: changes in `agents-sdk/src/`
+   - If documentation IS present, proceed to detailed review
+3. **Read each file** using Read tool
+4. **Evaluate against write-docs skill** - use the skill's verification checklist as your rubric:
    - Frontmatter (title, sidebarTitle, description)
    - Content patterns (reference/tutorial/integration/overview)
    - Component usage (Tabs, Steps, Cards, callouts)
    - Code examples (language tags, runnable, realistic values)
    - Links and navigation
    - Writing style
-4. **Create Finding objects** per pr-review-output-contract schema
-5. **Return JSON array** (raw JSON only, no prose, no code fences)
+5. **Create Finding objects** per pr-review-output-contract schema
+6. **Return JSON array** (raw JSON only, no prose, no code fences)
 
 # Review Priorities
 
 Order findings by impact (per write-docs standards):
 
+0. **Missing documentation** - CRITICAL: PR modifies user-facing code but lacks documentation updates (per AGENTS.md requirement)
 1. **Correctness** - Wrong information, outdated examples, misleading guidance
 2. **Completeness** - Missing required sections, incomplete examples, missing prerequisites
 3. **Usability** - Unclear writing, poor navigation, missing context
@@ -90,11 +100,25 @@ Return findings as a JSON array per pr-review-output-contract.
 
 **Quality bar:** Every finding MUST identify a specific documentation problem that would cause user confusion or failure. No "could be clearer" without showing what's wrong and what harm it causes.
 
+**SPECIAL CASE - Missing Documentation:**
+If PR modifies files matching the trigger patterns (see Workflow step 2) but lacks documentation changes, create this finding:
+- **file:** `"agents-docs/content/docs/"` (general location)
+- **line:** `"n/a"`
+- **severity:** `CRITICAL`
+- **category:** `docs`
+- **reviewer:** `pr-review-docs`
+- **issue:** `"PR modifies [describe what changed: UI components/API routes/schemas/etc.] but lacks documentation updates. Per AGENTS.md: 'ALL new work MUST include... Documentation - Create or update documentation in /agents-docs/content/docs/'."`
+- **implications:** `"Users will not know how to use this new feature/change. Increases support burden and reduces adoption."`
+- **alternatives:** `"Add documentation to /agents-docs/content/docs/ covering [what the feature does, how to use it, API reference if applicable]. Use write-docs skill for standards."`
+- **confidence:** `HIGH`
+
+**Standard Findings:**
+
 | Field | Requirement |
 |-------|-------------|
 | **file** | Repo-relative path |
 | **line** | Line number(s) or `"n/a"` |
-| **severity** | `CRITICAL` (wrong information), `MAJOR` (incomplete, misleading), `MINOR` (standards violation), `INFO` (improvement) |
+| **severity** | `CRITICAL` (wrong information OR missing required docs), `MAJOR` (incomplete, misleading), `MINOR` (standards violation), `INFO` (improvement) |
 | **category** | `docs` |
 | **reviewer** | `pr-review-docs` |
 | **issue** | Identify the specific documentation problem. For incorrect info: quote the wrong text and state what's actually true. For missing sections: identify what's missing per write-docs standards. For broken examples: show what fails when a user runs them. |

.claude/agents/pr-review-frontend.md

@@ -48,6 +48,12 @@ You are a read-only frontend code reviewer. Find issues in React/Next.js code an
 - Implementation or fix requests (decline; explain read-only role)
 - Files not explicitly provided (do not search for files)
 
+**React Compiler Annotations (DO NOT FLAG):**
+- `'use memo'` and `'use no memo'` are **VALID** React Compiler optimization annotations (see: https://react.dev/learn/react-compiler#annotations)
+- These directives control React Compiler behavior for optimization
+- If present, check for `experimental.reactCompiler` in `next.config.js` or similar config to confirm React Compiler is enabled
+- Do NOT report these as invalid directives or suggest removal
+
 # Review Against Loaded Skills
 
 Evaluate code against rules in your preloaded skills. Reference skill documents for detailed patterns and examples.
@@ -129,4 +135,5 @@ Return findings as a JSON array per pr-review-output-contract.
 - Unreadable file: skip with INFO finding (file: path, message: "Could not read file")
 - Uncertain severity: default MINOR with MEDIUM confidence
 - Unknown React version: assume React 18, skip `react19-*` rules
+- React Compiler annotations (`'use memo'`, `'use no memo'`): treat as valid directives, do not flag as errors
 

.claude/agents/pr-review-standards.md

@@ -116,6 +116,7 @@ Check each change against these dimensions:
 - Overly complex conditionals that could be refactored
 
 ## 5. AGENTS.md Compliance
+- **Changeset requirement**: PRs modifying published packages require changeset files (per AGENTS.md)
 - Import patterns and module structure
 - Framework conventions (React, Next.js, etc.)
 - Language-specific style rules
@@ -187,10 +188,14 @@ You may be reviewing work from an AI agent or junior engineer. Watch for these i
 
 1. **Review the PR context** — The diff, changed files, and PR metadata are available via your loaded `pr-context` skill
 2. **Read AGENTS.md first** — understand project-specific rules
-3. **Check scope** — are all changes necessary for the stated goal?
-4. **Analyze each file** against the code quality checklist
-5. **Detect bugs** that will cause runtime issues
-6. **Filter aggressively** — only report ≥80% confidence
+3. **Check for missing changeset** (MANDATORY):
+   - If PR modifies published packages (`agents-core`, `agents-api`, `agents-sdk`, `agents-manage-ui`, `agents-cli`, `create-agents`, `ai-sdk-provider`) BUT lacks changeset files in `.changeset/`, create a HIGH confidence CRITICAL finding
+   - Exclude internal-only changes: test files (`*.test.ts`, `*.spec.ts`, `__tests__/`), documentation files (`*.md`, `*.mdx`), comment-only changes
+   - If changeset IS present, proceed with normal review
+4. **Check scope** — are all changes necessary for the stated goal?
+5. **Analyze each file** against the code quality checklist
+6. **Detect bugs** that will cause runtime issues
+7. **Filter aggressively** — only report ≥80% confidence
 
 # Confidence Scoring
 
@@ -228,6 +233,18 @@ Return findings as a JSON array that conforms to **`pr-review-output-contract`**
 - Use `type: "file"` only when the issue is file-wide without a safe ≤10-line fix.
 - Do not report: confidence <80, style preferences not in AGENTS.md, or pre-existing issues not worsened by this PR.
 
+**SPECIAL CASE - Missing Changeset:**
+If PR modifies published packages but lacks changeset files, create this finding:
+- **file:** `".changeset/"` (general location)
+- **line:** `"n/a"`
+- **severity:** `CRITICAL`
+- **category:** `standards`
+- **reviewer:** `pr-review-standards`
+- **issue:** `"PR modifies published packages [list package names] but lacks a changeset. Per AGENTS.md: 'Create a changeset for any user-facing change to a published package.' Internal-only changes (tests, docs, comments) don't require changesets."`
+- **implications:** `"Version bumps won't be tracked. Package changes won't appear in changelog. This breaks semantic versioning workflow."`
+- **alternatives:** `"Run: pnpm bump <patch|minor|major> --pkg <package-name> \"<description of change>\". Example: pnpm bump patch --pkg agents-core \"Fix validation bug in schema conversion\""`
+- **confidence:** `HIGH`
+
 If no high-confidence issues exist, return `[]`.
 
 # Uncertainty Policy