@schoolai/shipyard-mcp 0.2.2 → 0.3.0

package/apps/hook/dist/{index.js → index.cjs}

@@ -28488,13 +28488,459 @@ var require_undici = __commonJS({
 // src/index.ts
 init_cjs_shims();
 
+// ../../packages/shared/dist/instructions/index.mjs
+init_cjs_shims();
+var TOOL_NAMES = {
+  ADD_ARTIFACT: "add_artifact",
+  ADD_PR_REVIEW_COMMENT: "add_pr_review_comment",
+  COMPLETE_TASK: "complete_task",
+  CREATE_PLAN: "create_plan",
+  EXECUTE_CODE: "execute_code",
+  LINK_PR: "link_pr",
+  READ_PLAN: "read_plan",
+  REGENERATE_SESSION_TOKEN: "regenerate_session_token",
+  REQUEST_USER_INPUT: "request_user_input",
+  SETUP_REVIEW_NOTIFICATION: "setup_review_notification",
+  UPDATE_BLOCK_CONTENT: "update_block_content",
+  UPDATE_PLAN: "update_plan"
+};
+var CRITICAL_USAGE_SECTION = `## CRITICAL: When to Use Shipyard
+
+\u26A0\uFE0F **MANDATORY USAGE RULE:**
+
+If Shipyard is available, you MUST use it for:
+- Creating implementation plans
+- Tracking work that needs human review
+- Documenting proof of work (screenshots, videos, test results)
+- ANY user request involving "plan", "track", "verify", or "prove"
+
+**DO NOT:**
+- Create plans manually in chat or as markdown files
+- Write implementation docs yourself when Shipyard is available
+- Suggest alternatives to Shipyard for trackable work
+- Overthink whether to use it - WHEN IN DOUBT, USE SHIPYARD
+
+**Decision Tree:**
+\`\`\`
+Need to create/track/verify work?
+        \u2502
+        \u25BC
+  Shipyard available?
+     \u2502           \u2502
+    YES          NO
+     \u2502           \u2502
+     \u25BC           \u25BC
+  USE IT     Manual approach
+  NOW        (tell user why)
+\`\`\``;
+var DELIVERABLES_SECTION = `## What are Deliverables?
+
+Deliverables are measurable outcomes you can **prove** with artifacts (screenshots, videos, test results).
+
+**Good deliverables (provable):**
+\`\`\`
+- [ ] Screenshot of working login page {#deliverable}
+- [ ] Video showing drag-and-drop feature {#deliverable}
+- [ ] Test results showing all tests pass {#deliverable}
+\`\`\`
+
+**Bad deliverables (not provable - these are tasks, not deliverables):**
+\`\`\`
+- [ ] Implement getUserMedia API  \u2190 Implementation detail, not provable
+- [ ] Add error handling          \u2190 Can't capture this with an artifact
+- [ ] Refactor authentication     \u2190 Too vague, no visual proof
+\`\`\`
+
+**Rule:** If you can't screenshot/record/export it, it's not a deliverable.`;
+var ARTIFACT_TYPES_SECTION = `## Artifact Types
+
+| Type | Use For | File Formats |
+|------|---------|--------------|
+| \`screenshot\` | UI changes, visual proof, error states | .png, .jpg, .webp |
+| \`video\` | Complex flows, interactions, animations | .mp4, .webm |
+| \`test_results\` | Test output, coverage reports | .json, .txt, .xml |
+| \`diff\` | Code changes, before/after comparisons | .diff, .patch |`;
+var TIPS_SECTION = `## Tips for Effective Use
+
+1. **Plan deliverables first** - Decide what proves success before coding
+2. **Capture during work** - Take screenshots as you implement, not after
+3. **Be specific** - "Login page with error state" beats "Screenshot"
+4. **Link every artifact** - Always set \`deliverableId\` for auto-completion
+5. **Check feedback** - Read reviewer comments and iterate`;
+var WHEN_NOT_TO_USE_SECTION = `## When NOT to Use Shipyard
+
+Skip Shipyard for:
+- Quick answers or research questions (no artifacts to capture)
+- Internal refactoring with no visible output
+- Tasks where proof adds no value (trivial fixes)
+- Exploration or debugging sessions
+- Pure documentation without implementation`;
+var USER_INPUT_SECTION = `## Human-Agent Communication
+
+**\`${TOOL_NAMES.REQUEST_USER_INPUT}\` is THE primary way to communicate with humans during active work.**
+
+Shipyard is the central hub where humans manage AI agents. When you need to ask a question, get clarification, or request a decision - use \`${TOOL_NAMES.REQUEST_USER_INPUT}\`. The human is already in the browser viewing your plan. That's where conversations should happen.
+
+### Why Use ${TOOL_NAMES.REQUEST_USER_INPUT}
+
+- **Context:** The human sees your question alongside the plan, artifacts, and comments
+- **History:** All exchanges are logged in the plan's activity feed
+- **Continuity:** The conversation stays attached to the work, not scattered across chat windows
+- **Flexibility:** 8 input types, multi-question forms, "Other" escape hatch for custom answers
+
+### Replace Platform Tools
+
+**ALWAYS prefer \`${TOOL_NAMES.REQUEST_USER_INPUT}\` over platform-specific tools:**
+
+| Platform | DON'T Use | Use Instead |
+|----------|-----------|-------------|
+| Claude Code | \`AskUserQuestion\` | \`${TOOL_NAMES.REQUEST_USER_INPUT}\` |
+| Cursor | Built-in prompts | \`${TOOL_NAMES.REQUEST_USER_INPUT}\` |
+| Windsurf | Native dialogs | \`${TOOL_NAMES.REQUEST_USER_INPUT}\` |
+| Claude Desktop | Chat questions | \`${TOOL_NAMES.REQUEST_USER_INPUT}\` |
+
+### When to Ask
+
+Use \`${TOOL_NAMES.REQUEST_USER_INPUT}\` when you need:
+- Clarification on requirements ("Which auth provider?")
+- Decisions that affect implementation ("PostgreSQL or SQLite?")
+- Confirmation before destructive actions ("Delete this file?")
+- User preferences ("Rate this approach 1-5")
+- Any information you can't infer from context
+
+### Example
+
+\`\`\`typescript
+const result = await requestUserInput({
+  message: "Which database should we use?",
+  type: "choice",
+  options: ["PostgreSQL", "SQLite", "MongoDB"],
+  timeout: 600  // 10 minutes
+});
+
+if (result.success) {
+  console.log("User chose:", result.response);
+}
+\`\`\`
+
+**Note:** The MCP tool is named \`${TOOL_NAMES.REQUEST_USER_INPUT}\` (snake_case). Inside \`${TOOL_NAMES.EXECUTE_CODE}\`, it's available as \`requestUserInput()\` (camelCase).`;
+var TROUBLESHOOTING_SECTION = `## Troubleshooting
+
+**Browser doesn't open:** Check MCP server is running and accessible.
+
+**Upload fails:** Verify file path exists. For GitHub uploads, check \`GITHUB_TOKEN\` has repo write access.
+
+**No auto-complete:** Ensure every deliverable has an artifact with matching \`deliverableId\`.
+
+**Plan not syncing:** Check WebSocket connection to registry server.
+
+**Input request times out:** User may not have seen it or needs more time. Default timeout is 30 minutes. Try again with a longer timeout or rephrase the question.
+
+**Input request declined:** User clicked "Decline." Rephrase your question, proceed with a reasonable default, or use a different approach.
+
+**No response to input:** Check if browser is connected to the plan. User may have closed the browser window.`;
+var COMMON_INSTRUCTIONS = [
+  CRITICAL_USAGE_SECTION,
+  USER_INPUT_SECTION,
+  DELIVERABLES_SECTION,
+  ARTIFACT_TYPES_SECTION,
+  TIPS_SECTION,
+  WHEN_NOT_TO_USE_SECTION,
+  TROUBLESHOOTING_SECTION
+].join("\n\n");
+var CLAUDE_CODE_HEADER = `[SHIPYARD] Collaborative planning with human review & proof-of-work tracking.`;
+var PLAN_MODE_WORKFLOW = `## How to Use (Claude Code with Hooks)
+
+You have the **full Shipyard experience** with automatic hooks. Use native plan mode:
+
+### Workflow
+
+1. **Enter plan mode** (Shift+Tab) \u2192 Browser opens with live plan automatically
+2. **Write your plan** with \`{#deliverable}\` markers for provable outcomes
+3. **Exit plan mode** \u2192 Hook **BLOCKS** until human approves or requests changes
+4. **On approval** \u2192 You automatically receive: planId, sessionToken, deliverable IDs
+5. **Do the work** \u2192 Take screenshots/videos as you implement
+6. **Upload artifacts** \u2192 \`${TOOL_NAMES.ADD_ARTIFACT}(filePath, deliverableId)\` for each deliverable
+7. **Auto-complete** \u2192 When all deliverables have artifacts, task completes with snapshot URL
+
+### After Approval
+
+You only need ONE tool: \`${TOOL_NAMES.ADD_ARTIFACT}\`
+
+The hook automatically injects everything you need (planId, sessionToken, deliverables).
+Just call \`${TOOL_NAMES.ADD_ARTIFACT}\` with the file path and deliverable ID.
+
+\`\`\`typescript
+// Example: After approval, you'll have these in context
+// planId: "abc123"
+// sessionToken: "xyz..."
+// deliverables: [{ id: "del_xxx", text: "Screenshot of login" }]
+
+await addArtifact({
+  planId,
+  sessionToken,
+  type: 'screenshot',
+  filename: 'login-page.png',
+  source: 'file',
+  filePath: '/tmp/screenshot.png',
+  deliverableId: deliverables[0].id
+});
+\`\`\`
+
+When the last deliverable gets an artifact, the task auto-completes and returns a snapshot URL.`;
+var IMPORTANT_NOTES = `## Important Notes for Claude Code
+
+- **DO NOT call \`createPlan()\` directly** - The hook handles plan creation when you enter plan mode
+- **DO NOT use the Shipyard skill** - The hook provides everything you need
+- **DO NOT poll for approval** - The hook blocks automatically until human decides
+- **DO use plan mode** for ANY work that needs tracking, verification, or human review
+- **DO use \`${TOOL_NAMES.REQUEST_USER_INPUT}\`** instead of \`AskUserQuestion\` - The human is in the browser viewing your plan, questions should appear there`;
+var CLAUDE_CODE_INSTRUCTIONS = [
+  CLAUDE_CODE_HEADER,
+  "",
+  CRITICAL_USAGE_SECTION,
+  "",
+  USER_INPUT_SECTION,
+  "",
+  PLAN_MODE_WORKFLOW,
+  "",
+  DELIVERABLES_SECTION,
+  "",
+  ARTIFACT_TYPES_SECTION,
+  "",
+  IMPORTANT_NOTES,
+  "",
+  TIPS_SECTION,
+  "",
+  WHEN_NOT_TO_USE_SECTION,
+  "",
+  TROUBLESHOOTING_SECTION
+].join("\n");
+var MCP_DIRECT_HEADER = `# Shipyard: Your Agent Management Hub
+
+> **Shipyard is the central interface where humans manage AI agents.** Plans, artifacts, feedback, and communication all happen here.
+
+Shipyard turns invisible agent work into reviewable, verifiable tasks. Instead of trusting that code was written correctly, reviewers see screenshots, videos, and test results as proof.
+
+**Key principle:** When you're working in Shipyard, ALL human-agent communication should happen through Shipyard's \`${TOOL_NAMES.REQUEST_USER_INPUT}\` tool. The human is already in the browser viewing your plan - that's where they expect to interact with you.`;
+var MCP_TOOLS_OVERVIEW = `## Available MCP Tools
+
+| Tool | Purpose |
+|------|---------|
+| \`${TOOL_NAMES.REQUEST_USER_INPUT}\` | **THE primary human-agent communication channel** - Ask questions, get decisions, request clarification |
+| \`${TOOL_NAMES.EXECUTE_CODE}\` | Run TypeScript that calls Shipyard APIs (recommended for multi-step operations) |
+
+### ${TOOL_NAMES.REQUEST_USER_INPUT}: Your Direct Line to the Human
+
+This is how you talk to humans during active work. Don't use your platform's built-in question tools (AskUserQuestion, etc.) - use this instead. The human is in the browser viewing your plan, and that's where they expect to see your questions.
+
+**Preferred approach:** Use \`${TOOL_NAMES.EXECUTE_CODE}\` to chain multiple API calls in one step.`;
+var MCP_WORKFLOW = `## Workflow (MCP Direct)
+
+### Step 1: Create Plan
+
+\`\`\`typescript
+const plan = await createPlan({
+  title: "Add user authentication",
+  content: \`
+## Deliverables
+- [ ] Screenshot of login page {#deliverable}
+- [ ] Screenshot of error handling {#deliverable}
+
+## Implementation
+1. Create login form component
+2. Add validation
+3. Connect to auth API
+\`
+});
+
+const { planId, sessionToken, deliverables, monitoringScript } = plan;
+// deliverables = [{ id: "del_xxx", text: "Screenshot of login page" }, ...]
+\`\`\`
+
+### Step 2: Wait for Approval
+
+For platforms without hooks, run the monitoring script in the background:
+
+\`\`\`bash
+# The monitoringScript polls for approval status
+# Run it in background while you wait
+bash <(echo "$monitoringScript") &
+\`\`\`
+
+Or poll manually:
+
+\`\`\`typescript
+const status = await readPlan(planId, sessionToken);
+if (status.status === "in_progress") {
+  // Approved! Proceed with work
+}
+if (status.status === "changes_requested") {
+  // Read feedback, make changes
+}
+\`\`\`
+
+### Step 3: Do the Work
+
+Implement the feature, taking screenshots/recordings as you go.
+
+### Step 4: Upload Artifacts
+
+\`\`\`typescript
+await addArtifact({
+  planId,
+  sessionToken,
+  type: 'screenshot',
+  filename: 'login-page.png',
+  source: 'file',
+  filePath: '/path/to/screenshot.png',
+  deliverableId: deliverables[0].id  // Links to specific deliverable
+});
+
+const result = await addArtifact({
+  planId,
+  sessionToken,
+  type: 'screenshot',
+  filename: 'error-handling.png',
+  source: 'file',
+  filePath: '/path/to/error.png',
+  deliverableId: deliverables[1].id
+});
+
+// Auto-complete triggers when ALL deliverables have artifacts
+if (result.allDeliverablesComplete) {
+  console.log('Done!', result.snapshotUrl);
+}
+\`\`\``;
+var API_REFERENCE = `## API Reference
+
+### createPlan(options)
+
+Creates a new plan and opens it in the browser.
+
+**Parameters:**
+- \`title\` (string) - Plan title
+- \`content\` (string) - Markdown content with \`{#deliverable}\` markers
+- \`repo\` (string, optional) - GitHub repo for artifact storage
+- \`prNumber\` (number, optional) - PR number to link
+
+**Returns:** \`{ planId, sessionToken, url, deliverables, monitoringScript }\`
+
+### readPlan(planId, sessionToken, options?)
+
+Reads current plan state.
+
+**Parameters:**
+- \`planId\` (string) - Plan ID
+- \`sessionToken\` (string) - Session token from createPlan
+- \`options.includeAnnotations\` (boolean) - Include reviewer comments
+
+**Returns:** \`{ content, status, title, deliverables }\`
+
+### addArtifact(options)
+
+Uploads proof-of-work artifact.
+
+**Parameters:**
+- \`planId\` (string) - Plan ID
+- \`sessionToken\` (string) - Session token
+- \`type\` ('screenshot' | 'video' | 'test_results' | 'diff')
+- \`filename\` (string) - File name
+- \`source\` ('file' | 'url' | 'base64')
+- \`filePath\` (string) - Local file path (when source='file')
+- \`deliverableId\` (string, optional) - Links artifact to deliverable
+
+**Returns:** \`{ artifactId, url, allDeliverablesComplete, snapshotUrl? }\`
+
+### requestUserInput(options)
+
+Asks user a question via browser modal.
+
+**Parameters:**
+- \`message\` (string) - Question to ask
+- \`type\` (string) - Input type (see below)
+- \`options\` (string[], for 'choice') - Available choices
+- \`timeout\` (number, optional) - Timeout in seconds
+- Type-specific parameters (min, max, format, etc.)
+
+**Returns:** \`{ success, response?, status }\`
+
+**Supported types (8 total):**
+1. \`text\` - Single-line text
+2. \`multiline\` - Multi-line text area
+3. \`choice\` - Radio/checkbox/dropdown (auto-adds "Other" option)
+   - Auto-switches: 1-8 options = radio/checkbox, 9+ = dropdown
+   - \`multiSelect: true\` for checkboxes
+   - \`displayAs: 'dropdown'\` to force dropdown UI
+4. \`confirm\` - Yes/No buttons
+5. \`number\` - Numeric input with validation
+   - \`min\`, \`max\`, \`format\` ('integer' | 'decimal' | 'currency' | 'percentage')
+6. \`email\` - Email validation
+   - \`domain\` for restriction
+7. \`date\` - Date picker with range
+   - \`minDate\`, \`maxDate\` (YYYY-MM-DD format)
+8. \`rating\` - Scale rating (auto-selects stars/numbers)
+   - \`min\`, \`max\`, \`style\` ('stars' | 'numbers' | 'emoji'), \`labels\`
+
+**Response format:**
+- All responses are strings
+- choice (single): \`"PostgreSQL"\` or custom text from "Other"
+- choice (multi): \`"option1, option2"\` (comma-space separated)
+- confirm: \`"yes"\` or \`"no"\` (lowercase)
+- number: \`"42"\` or \`"3.14"\`
+- email: \`"user@example.com"\`
+- date: \`"2026-01-24"\` (ISO 8601)
+- rating: \`"4"\` (integer as string)
+- See docs/INPUT-RESPONSE-FORMATS.md for complete specification`;
+var HANDLING_FEEDBACK = `## Handling Reviewer Feedback
+
+\`\`\`typescript
+const status = await readPlan(planId, sessionToken, {
+  includeAnnotations: true
+});
+
+if (status.status === "changes_requested") {
+  // Read the content for inline comments
+  console.log(status.content);
+
+  // Make changes based on feedback
+  // Upload new artifacts
+  // Plan will transition back to pending_review
+}
+\`\`\``;
+var MCP_DIRECT_INSTRUCTIONS = [
+  MCP_DIRECT_HEADER,
+  "",
+  CRITICAL_USAGE_SECTION,
+  "",
+  USER_INPUT_SECTION,
+  "",
+  MCP_TOOLS_OVERVIEW,
+  "",
+  MCP_WORKFLOW,
+  "",
+  DELIVERABLES_SECTION,
+  "",
+  ARTIFACT_TYPES_SECTION,
+  "",
+  API_REFERENCE,
+  "",
+  HANDLING_FEEDBACK,
+  "",
+  TIPS_SECTION,
+  "",
+  WHEN_NOT_TO_USE_SECTION,
+  "",
+  TROUBLESHOOTING_SECTION
+].join("\n");
+
 // src/adapters/claude-code.ts
 init_cjs_shims();
 
 // ../../packages/schema/dist/index.mjs
 init_cjs_shims();
 
-// ../../packages/schema/dist/yjs-helpers-BBG4tC2R.mjs
+// ../../packages/schema/dist/yjs-helpers-DxqhtEAu.mjs
 init_cjs_shims();
 
 // ../../packages/schema/dist/plan.mjs
@@ -42532,9 +42978,17 @@ var PlanEventSchema = external_exports.discriminatedUnion("type", [
         "text",
         "multiline",
         "choice",
-        "confirm"
+        "confirm",
+        "number",
+        "email",
+        "date",
+        "dropdown",
+        "rating",
+        "multi"
       ]),
-      requestMessage: external_exports.string()
+      requestMessage: external_exports.string().optional(),
+      questionCount: external_exports.number().optional(),
+      isBlocker: external_exports.boolean().optional()
     })
   }),
   PlanEventBaseSchema.extend({
@@ -42552,7 +43006,8 @@ var PlanEventSchema = external_exports.discriminatedUnion("type", [
   PlanEventBaseSchema.extend({
     type: external_exports.literal("agent_activity"),
     data: AgentActivityDataSchema
-  })
+  }),
+  PlanEventBaseSchema.extend({ type: external_exports.literal("session_token_regenerated") })
 ]);
 var PlanMetadataBaseSchema = external_exports.object({
   id: external_exports.string(),
@@ -42602,10 +43057,9 @@ var PlanMetadataSchema = external_exports.discriminatedUnion("status", [
 var BaseArtifactSchema = external_exports.object({
   id: external_exports.string(),
   type: external_exports.enum([
-    "screenshot",
-    "video",
-    "test_results",
-    "diff"
+    "html",
+    "image",
+    "video"
   ]),
   filename: external_exports.string(),
   description: external_exports.string().optional(),
@@ -42665,7 +43119,7 @@ var PRReviewCommentSchema = external_exports.object({
   resolved: external_exports.boolean().optional()
 });
 
-// ../../packages/schema/dist/yjs-helpers-BBG4tC2R.mjs
+// ../../packages/schema/dist/yjs-helpers-DxqhtEAu.mjs
 function assertNever2(value) {
   throw new Error(`Unhandled discriminated union member: ${JSON.stringify(value)}`);
 }
@@ -42770,26 +43224,161 @@ var InputRequestBaseSchema = external_exports.object({
   message: external_exports.string().min(1, "Message cannot be empty"),
   status: external_exports.enum(InputRequestStatusValues),
   defaultValue: external_exports.string().optional(),
-  timeout: external_exports.number().int().min(10, "Timeout must be at least 10 seconds").max(14400, "Timeout cannot exceed 4 hours").optional(),
+  timeout: external_exports.number().int().min(300, "Timeout must be at least 5 minutes (300 seconds)").max(1800, "Timeout cannot exceed 30 minutes (1800 seconds)").optional(),
   planId: external_exports.string().optional(),
   response: external_exports.unknown().optional(),
   answeredAt: external_exports.number().optional(),
-  answeredBy: external_exports.string().optional()
+  answeredBy: external_exports.string().optional(),
+  isBlocker: external_exports.boolean().optional()
 });
 var TextInputSchema = InputRequestBaseSchema.extend({ type: external_exports.literal("text") });
 var MultilineInputSchema = InputRequestBaseSchema.extend({ type: external_exports.literal("multiline") });
+var ChoiceOptionSchema = external_exports.union([external_exports.string(), external_exports.object({
+  value: external_exports.string(),
+  label: external_exports.string().optional(),
+  description: external_exports.string().optional(),
+  icon: external_exports.string().optional(),
+  disabled: external_exports.boolean().optional()
+})]);
 var ChoiceInputSchema = InputRequestBaseSchema.extend({
   type: external_exports.literal("choice"),
-  options: external_exports.array(external_exports.string()).min(1, "Choice requests must have at least one option"),
-  multiSelect: external_exports.boolean().optional()
+  options: external_exports.array(ChoiceOptionSchema).min(1, "Choice requests must have at least one option"),
+  multiSelect: external_exports.boolean().optional(),
+  displayAs: external_exports.enum([
+    "radio",
+    "checkbox",
+    "dropdown"
+  ]).optional(),
+  placeholder: external_exports.string().optional()
 });
 var ConfirmInputSchema = InputRequestBaseSchema.extend({ type: external_exports.literal("confirm") });
+var NumberInputSchema = InputRequestBaseSchema.extend({
+  type: external_exports.literal("number"),
+  min: external_exports.number().optional(),
+  max: external_exports.number().optional(),
+  format: external_exports.enum([
+    "integer",
+    "decimal",
+    "currency",
+    "percentage"
+  ]).optional()
+}).refine((data) => data.min === void 0 || data.max === void 0 || data.min <= data.max, { message: "min must be <= max" });
+var EmailInputSchema = InputRequestBaseSchema.extend({
+  type: external_exports.literal("email"),
+  domain: external_exports.string().optional()
+});
+var DateInputSchema = InputRequestBaseSchema.extend({
+  type: external_exports.literal("date"),
+  min: external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/, "Date must be in YYYY-MM-DD format").optional(),
+  max: external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/, "Date must be in YYYY-MM-DD format").optional()
+}).refine((data) => data.min === void 0 || data.max === void 0 || new Date(data.min) <= new Date(data.max), { message: "min date must be before or equal to max date" });
+var RatingInputSchema = InputRequestBaseSchema.extend({
+  type: external_exports.literal("rating"),
+  min: external_exports.number().int().optional(),
+  max: external_exports.number().int().optional(),
+  style: external_exports.enum([
+    "stars",
+    "numbers",
+    "emoji"
+  ]).optional(),
+  labels: external_exports.object({
+    low: external_exports.string().optional(),
+    high: external_exports.string().optional()
+  }).optional()
+}).refine((data) => {
+  if (data.min === void 0 || data.max === void 0) return true;
+  return data.min <= data.max && data.max - data.min <= 20;
+}, { message: "Rating scale must have min <= max and at most 20 items" });
 var InputRequestSchema = external_exports.discriminatedUnion("type", [
   TextInputSchema,
   MultilineInputSchema,
   ChoiceInputSchema,
-  ConfirmInputSchema
+  ConfirmInputSchema,
+  NumberInputSchema,
+  EmailInputSchema,
+  DateInputSchema,
+  RatingInputSchema
 ]);
+var MAX_QUESTIONS_PER_REQUEST = 10;
+var QuestionBaseSchema = external_exports.object({
+  message: external_exports.string().min(1, "Message cannot be empty"),
+  defaultValue: external_exports.string().optional()
+});
+var TextQuestionSchema = QuestionBaseSchema.extend({ type: external_exports.literal("text") });
+var MultilineQuestionSchema = QuestionBaseSchema.extend({ type: external_exports.literal("multiline") });
+var ChoiceQuestionSchema = QuestionBaseSchema.extend({
+  type: external_exports.literal("choice"),
+  options: external_exports.array(ChoiceOptionSchema).min(1, "Choice questions must have at least one option"),
+  multiSelect: external_exports.boolean().optional(),
+  displayAs: external_exports.enum([
+    "radio",
+    "checkbox",
+    "dropdown"
+  ]).optional(),
+  placeholder: external_exports.string().optional()
+});
+var ConfirmQuestionSchema = QuestionBaseSchema.extend({ type: external_exports.literal("confirm") });
+var NumberQuestionSchema = QuestionBaseSchema.extend({
+  type: external_exports.literal("number"),
+  min: external_exports.number().optional(),
+  max: external_exports.number().optional(),
+  format: external_exports.enum([
+    "integer",
+    "decimal",
+    "currency",
+    "percentage"
+  ]).optional()
+}).refine((data) => data.min === void 0 || data.max === void 0 || data.min <= data.max, { message: "min must be <= max" });
+var EmailQuestionSchema = QuestionBaseSchema.extend({
+  type: external_exports.literal("email"),
+  domain: external_exports.string().optional()
+});
+var DateQuestionSchema = QuestionBaseSchema.extend({
+  type: external_exports.literal("date"),
+  min: external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/, "Date must be in YYYY-MM-DD format").optional(),
+  max: external_exports.string().regex(/^\d{4}-\d{2}-\d{2}$/, "Date must be in YYYY-MM-DD format").optional()
+}).refine((data) => data.min === void 0 || data.max === void 0 || new Date(data.min) <= new Date(data.max), { message: "min date must be before or equal to max date" });
+var RatingQuestionSchema = QuestionBaseSchema.extend({
+  type: external_exports.literal("rating"),
+  min: external_exports.number().int().optional(),
+  max: external_exports.number().int().optional(),
+  style: external_exports.enum([
+    "stars",
+    "numbers",
+    "emoji"
+  ]).optional(),
+  labels: external_exports.object({
+    low: external_exports.string().optional(),
+    high: external_exports.string().optional()
+  }).optional()
+}).refine((data) => {
+  if (data.min === void 0 || data.max === void 0) return true;
+  return data.min <= data.max && data.max - data.min <= 20;
+}, { message: "Rating scale must have min <= max and at most 20 items" });
+var QuestionSchema = external_exports.discriminatedUnion("type", [
+  TextQuestionSchema,
+  MultilineQuestionSchema,
+  ChoiceQuestionSchema,
+  ConfirmQuestionSchema,
+  NumberQuestionSchema,
+  EmailQuestionSchema,
+  DateQuestionSchema,
+  RatingQuestionSchema
+]);
+var MultiQuestionInputRequestSchema = external_exports.object({
+  id: external_exports.string(),
+  createdAt: external_exports.number(),
+  type: external_exports.literal("multi"),
+  questions: external_exports.array(QuestionSchema).min(1, "At least one question is required").max(MAX_QUESTIONS_PER_REQUEST, `Maximum ${MAX_QUESTIONS_PER_REQUEST} questions allowed (8 recommended for optimal UX)`),
+  status: external_exports.enum(InputRequestStatusValues),
+  timeout: external_exports.number().int().min(10, "Timeout must be at least 10 seconds").max(14400, "Timeout cannot exceed 4 hours").optional(),
+  planId: external_exports.string().optional(),
+  responses: external_exports.record(external_exports.string(), external_exports.unknown()).optional(),
+  answeredAt: external_exports.number().optional(),
+  answeredBy: external_exports.string().optional(),
+  isBlocker: external_exports.boolean().optional()
+});
+var AnyInputRequestSchema = external_exports.union([InputRequestSchema, MultiQuestionInputRequestSchema]);
 var YDOC_KEYS = {
   METADATA: "metadata",
   DOCUMENT_FRAGMENT: "document",
@@ -43792,6 +44381,29 @@ var GitHubPRResponseSchema = external_exports.object({
   merged: external_exports.boolean(),
   head: external_exports.object({ ref: external_exports.string() })
 });
+var ROUTES = {
+  REGISTRY_LIST: "/registry",
+  REGISTRY_REGISTER: "/register",
+  REGISTRY_UNREGISTER: "/unregister",
+  PLAN_STATUS: (planId) => `/api/plan/${planId}/status`,
+  PLAN_HAS_CONNECTIONS: (planId) => `/api/plan/${planId}/has-connections`,
+  PLAN_TRANSCRIPT: (planId) => `/api/plan/${planId}/transcript`,
+  PLAN_SUBSCRIBE: (planId) => `/api/plan/${planId}/subscribe`,
+  PLAN_CHANGES: (planId) => `/api/plan/${planId}/changes`,
+  PLAN_UNSUBSCRIBE: (planId) => `/api/plan/${planId}/unsubscribe`,
+  PLAN_PR_DIFF: (planId, prNumber) => `/api/plans/${planId}/pr-diff/${prNumber}`,
+  PLAN_PR_FILES: (planId, prNumber) => `/api/plans/${planId}/pr-files/${prNumber}`,
+  HOOK_SESSION: "/api/hook/session",
+  HOOK_CONTENT: (planId) => `/api/hook/plan/${planId}/content`,
+  HOOK_REVIEW: (planId) => `/api/hook/plan/${planId}/review`,
+  HOOK_SESSION_TOKEN: (planId) => `/api/hook/plan/${planId}/session-token`,
+  HOOK_PRESENCE: (planId) => `/api/hook/plan/${planId}/presence`,
+  CONVERSATION_IMPORT: "/api/conversation/import",
+  WEB_TASK: (planId) => `/task/${planId}`
+};
+function createPlanWebUrl(baseUrl, planId) {
+  return `${baseUrl.replace(/\/$/, "")}${ROUTES.WEB_TASK(planId)}`;
+}
 var InviteTokenSchema = external_exports.object({
   id: external_exports.string(),
   tokenHash: external_exports.string(),
@@ -43809,6 +44421,42 @@ var InviteRedemptionSchema = external_exports.object({
   redeemedAt: external_exports.number(),
   tokenId: external_exports.string()
 });
+var GitFileStatusSchema = external_exports.enum([
+  "added",
+  "modified",
+  "deleted",
+  "renamed",
+  "copied",
+  "untracked"
+]);
+var LocalFileChangeSchema = external_exports.object({
+  path: external_exports.string(),
+  status: GitFileStatusSchema,
+  additions: external_exports.number(),
+  deletions: external_exports.number(),
+  patch: external_exports.string().optional()
+});
+var LocalChangesResponseSchema = external_exports.object({
+  available: external_exports.literal(true),
+  branch: external_exports.string(),
+  baseBranch: external_exports.string(),
+  staged: external_exports.array(LocalFileChangeSchema),
+  unstaged: external_exports.array(LocalFileChangeSchema),
+  untracked: external_exports.array(external_exports.string()),
+  files: external_exports.array(LocalFileChangeSchema)
+});
+var LocalChangesUnavailableReasonSchema = external_exports.enum([
+  "no_cwd",
+  "not_git_repo",
+  "mcp_not_connected",
+  "git_error"
+]);
+var LocalChangesUnavailableSchema = external_exports.object({
+  available: external_exports.literal(false),
+  reason: LocalChangesUnavailableReasonSchema,
+  message: external_exports.string()
+});
+var LocalChangesResultSchema = external_exports.discriminatedUnion("available", [LocalChangesResponseSchema, LocalChangesUnavailableSchema]);
 var ConversationExportStartMetaSchema = external_exports.object({
   exportId: external_exports.string(),
   totalChunks: external_exports.number().int().positive(),
@@ -43882,6 +44530,20 @@ function truncate(text, maxLength) {
   if (cleaned.length <= maxLength) return cleaned;
   return `${cleaned.slice(0, maxLength)}...`;
 }
+var TOOL_NAMES2 = {
+  ADD_ARTIFACT: "add_artifact",
+  ADD_PR_REVIEW_COMMENT: "add_pr_review_comment",
+  COMPLETE_TASK: "complete_task",
+  CREATE_PLAN: "create_plan",
+  EXECUTE_CODE: "execute_code",
+  LINK_PR: "link_pr",
+  READ_PLAN: "read_plan",
+  REGENERATE_SESSION_TOKEN: "regenerate_session_token",
+  REQUEST_USER_INPUT: "request_user_input",
+  SETUP_REVIEW_NOTIFICATION: "setup_review_notification",
+  UPDATE_BLOCK_CONTENT: "update_block_content",
+  UPDATE_PLAN: "update_plan"
+};
 var PlanIdSchema = external_exports.object({ planId: external_exports.string().min(1) });
 var PlanStatusResponseSchema = external_exports.object({ status: external_exports.string() });
 var HasConnectionsResponseSchema = external_exports.object({ hasConnections: external_exports.boolean() });
@@ -44004,6 +44666,41 @@ var planRouter = router({
   }),
   hasConnections: publicProcedure.input(PlanIdSchema).output(HasConnectionsResponseSchema).query(async ({ input, ctx }) => {
     return { hasConnections: await ctx.getPlanStore().hasActiveConnections(input.planId) };
+  }),
+  getLocalChanges: publicProcedure.input(PlanIdSchema).output(LocalChangesResultSchema).query(async ({ input, ctx }) => {
+    const metadata = getPlanMetadata(await ctx.getOrCreateDoc(input.planId));
+    if (!metadata) throw new TRPCError({
+      code: "NOT_FOUND",
+      message: "Plan not found"
+    });
+    const origin = metadata.origin;
+    const cwd = origin?.platform === "claude-code" ? origin.cwd : void 0;
+    if (!cwd) return {
+      available: false,
+      reason: "no_cwd",
+      message: "Plan has no associated working directory. Local changes are only available for plans created with working directory metadata."
+    };
+    return ctx.getLocalChanges(cwd);
+  }),
+  getFileContent: publicProcedure.input(external_exports.object({
+    planId: PlanIdSchema.shape.planId,
+    filePath: external_exports.string()
+  })).output(external_exports.object({
+    content: external_exports.string().nullable(),
+    error: external_exports.string().optional()
+  })).query(async ({ input, ctx }) => {
+    const metadata = getPlanMetadata(await ctx.getOrCreateDoc(input.planId));
+    if (!metadata) throw new TRPCError({
+      code: "NOT_FOUND",
+      message: "Plan not found"
+    });
+    const origin = metadata.origin;
+    const cwd = origin?.platform === "claude-code" ? origin.cwd : void 0;
+    if (!cwd) return {
+      content: null,
+      error: "No working directory available"
+    };
+    return ctx.getFileContent(cwd, input.filePath);
   })
 });
 var subscriptionRouter = router({
@@ -44046,9 +44743,6 @@ var CLAUDE_TOOL_NAMES = {
   EXIT_PLAN_MODE: "ExitPlanMode",
   ASK_USER_QUESTION: "AskUserQuestion"
 };
-var MCP_TOOL_NAMES = {
-  REQUEST_USER_INPUT: "request_user_input"
-};
 var CLAUDE_PERMISSION_MODES = {
   PLAN: "plan",
   DEFAULT: "default",
@@ -44145,7 +44839,7 @@ function handlePreToolUse(input) {
     );
     return {
       type: "tool_deny",
-      reason: `BLOCKED: Use the ${MCP_TOOL_NAMES.REQUEST_USER_INPUT} MCP tool instead for consistent browser UI. See the tool description for available parameters.`
+      reason: `BLOCKED: Use the ${TOOL_NAMES2.REQUEST_USER_INPUT} MCP tool instead. The human is in the browser viewing your plan - that's where they expect to interact with you. See the tool description for input types and parameters.`
     };
   }
   return { type: "passthrough" };
@@ -45749,7 +46443,7 @@ async function handleUpdatedPlanReview(sessionId, planId, planContent, _originMe
   }
   const baseUrl = webConfig.SHIPYARD_WEB_URL;
   logger.info(
-    { planId, url: `${baseUrl}/plan/${planId}` },
+    { planId, url: createPlanWebUrl(baseUrl, planId) },
     "Content synced, browser already open. Waiting for server approval..."
   );
   const decision = await waitForReviewDecision(planId, "");
@@ -45906,7 +46600,7 @@ Reviewer comment: ${decision.reviewComment}` : "";
         allow: false,
         message: `Plan is pending review.
 
-Open: ${baseUrl}/plan/${planId}`,
+Open: ${createPlanWebUrl(baseUrl, planId)}`,
         planId
       };
     case "draft":
@@ -45914,7 +46608,7 @@ Open: ${baseUrl}/plan/${planId}`,
         allow: false,
         message: `Plan is still in draft.
 
-Submit for review at: ${baseUrl}/plan/${planId}`,
+Submit for review at: ${createPlanWebUrl(baseUrl, planId)}`,
         planId
       };
     case "in_progress":
@@ -46071,46 +46765,10 @@ async function readStdin() {
   return Buffer.concat(chunks).toString("utf-8");
 }
 function outputSessionStartContext() {
-  const context = `[SHIPYARD] Collaborative planning with human review & proof-of-work tracking.
-
-IMPORTANT: Use native plan mode (Shift+Tab) to create plans. The hook handles everything automatically.
-
-## What are Deliverables?
-
-Deliverables are measurable outcomes you can prove with artifacts (screenshots, videos, test results).
-
-Good deliverables (provable):
-\`\`\`
-- [ ] Screenshot of working login page {#deliverable}
-- [ ] Video showing feature in action {#deliverable}
-- [ ] Test results showing all tests pass {#deliverable}
-\`\`\`
-
-Bad deliverables (implementation details, not provable):
-\`\`\`
-- [ ] Implement getUserMedia API  \u2190 This is a task, not a deliverable
-- [ ] Add error handling          \u2190 Can't prove this with an artifact
-\`\`\`
-
-## Workflow
-
-1. Enter plan mode (Shift+Tab) \u2192 Browser opens with live plan
-2. Write plan with {#deliverable} markers for provable outcomes
-3. Exit plan mode \u2192 Hook BLOCKS until human approves
-4. On approval \u2192 You receive planId, sessionToken, and deliverable IDs
-5. Do work \u2192 Take screenshots/videos as you go
-6. \`add_artifact(filePath, deliverableId)\` for each deliverable
-7. When all deliverables fulfilled \u2192 Auto-completes with snapshot URL
-
-## After Approval
-
-You only need ONE tool: \`add_artifact\`
-
-When the last deliverable gets an artifact, the task auto-completes and returns a snapshot URL for your PR.`;
   const hookOutput = {
     hookSpecificOutput: {
       hookEventName: "SessionStart",
-      additionalContext: context
+      additionalContext: CLAUDE_CODE_INSTRUCTIONS
     }
   };
   console.log(JSON.stringify(hookOutput));