feat: add shipping score delta, weekly streak persistence, and AI context for performance tracking

Author: Bharath-code

README.md

@@ -81,6 +81,7 @@ Five questions. AI scoring. A brief that stings when it should.
 - **Overall score** — averaged anchor at a glance
 - Riskiest assumption named explicitly
 - One async validation action you can do tonight
+- **Sunday Reminder**: Provides a helpful hint to set a recurring calendar event or cron job to close your week
 
 ### `loopkit track` — zero-overhead task management
 
@@ -96,22 +97,27 @@ Five questions. AI scoring. A brief that stings when it should.
 - Reads your brief + completed tasks for AI context
 - Generates drafts for Show HN, Twitter, and Indie Hackers
 - Per-draft: **use**, **edit in `$EDITOR`**, **regenerate**, or **skip**
-- Pre-launch checklist catches what you forgot
+- Pre-launch checklist catches what you forgot (runs *after* draft generation to get to value faster)
+- AI output remains calibrated by truncating long descriptions
+- Auto-triggers `celebrate` confetti, but only if you actually use a draft
 - Ship log saved to `.loopkit/ships/YYYY-MM-DD.md`
 
 ### `loopkit pulse` — feedback that comes to you
 
 - **Hosted Form & Widget**: Collect feedback async via your own `/pulse/[projectId]` hosted page or an embeddable JS widget on your landing page. Fully supports offline submission with automatic real-time sync.
-- Add responses from the CLI: `loopkit pulse --add "user said this"`
+- Add responses from the CLI: `loopkit pulse --add "user said this"` (automatically deduplicates)
+- Embeddable widget displays your actual `projectId` for easy copy-pasting
 - AI clusters responses into: Fix now / Validate later / Noise
 - Requires 5+ responses for clustering; shows raw list below threshold
 - Tag clusters directly to your sprint
 
 ### `loopkit loop` — the Sunday ritual
 
-- Aggregates your week: tasks, feedback, ship logs
-- Tracks your consecutive **Loop Streak** (e.g. `🔥 4 consecutive weeks`)
+- Aggregates your week: tasks, ship logs, **pulse feedback**, and **previous weeks' context**
+- Tracks your consecutive **Loop Streak** (e.g. `🔥 4 consecutive weeks`) and acknowledges streak breaks
 - AI recommends the single highest-leverage thing for next week
+- Smart Unstuck Mode: kicks in if you have 0 tasks, but only after your first week
+- Context-aware upgrade prompts at week 4 or 8
 - Override rate tracked — warns if you're overriding AI > 50% of the time
 - Generates a build-in-public post (280-char checked)
 - One decision. One post. Week done.

docs/USER_GUIDE.md

@@ -100,7 +100,9 @@ LoopKit replaces your scattered task manager with a simple `tasks.md` file:
 
 - **Auto-close via git:** Write `[#1]` in your commit message → task automatically marks as done
 - **Shipping score:** Done tasks / Total tasks × 100. Aim for > 70%.
+- **Score delta:** Compare your shipping score to last week (e.g. `↑+12%` or `↓-8%`) for a daily progress anchor.
 - **Stale task detection:** Tasks open for 3+ days get flagged. Keep, snooze, or cut them.
+- **Smart Benchmarks:** Compares your performance against the network, unlocking after you reach 4 weeks of data.
 - **Never deleted:** Cut tasks move to `cut.md` — you might want them later.
 
 **Why this matters:** Visibility creates accountability. A 60% shipping score is a data point, not a feeling.
@@ -120,12 +122,13 @@ Shipping isn't just deploying code. It's announcing it to the right people on th
 LoopKit asks: "What did you ship?" Then AI generates:
 
 - **Hacker News:** Title + body (no hype, developer tone)
-- **Twitter:** 3-tweet thread (build in public style)
+- **Twitter:** 3-tweet thread (build in public style, with character count warnings)
 - **Indie Hackers:** Narrative post (story-driven)
 
 You can use, edit, regenerate, or skip each draft.
+After reviewing the drafts, you run through a pre-launch checklist to ensure you updated your README, landing page, analytics, and feedback form.
 
-**Why this matters:** Writing launch copy takes 1-2 hours. LoopKit does it in 30 seconds. The faster you launch, the faster you learn.
+**Why this matters:** Writing launch copy takes 1-2 hours. LoopKit does it in 30 seconds. The faster you launch, the faster you learn. By generating the drafts first, you get immediate value before doing the housekeeping tasks.
 
 **Output:** Published posts + a `ship log` saved for your records.
 
@@ -141,8 +144,8 @@ Most founders build in a vacuum. LoopKit makes feedback collection automatic.
 
 **Two ways to collect:**
 
-1. **Inline:** `loopkit pulse --add "User said the pricing is confusing"`
-2. **Shareable form:** `loopkit pulse --share` → get a URL + QR code → embed on your site
+1. **Inline:** `loopkit pulse --add "User said the pricing is confusing"` (automatically deduplicates)
+2. **Shareable form:** `loopkit pulse --share` → get a URL + QR code → embed on your site (widget outputs your actual `projectId`)
 
 Once you have 5+ responses, AI clusters them into:
 
@@ -253,7 +256,7 @@ Your shipping score is `done tasks / total tasks × 100`.
 
 Consecutive weeks where you run `loopkit loop` and close your synthesis.
 
-A 4-week streak means you've shipped and reflected for a month straight. This is the #1 predictor of eventual success.
+If you miss a week, your streak resets, but you get a gentle acknowledgment to help you start a new streak today. A 4-week streak means you've shipped and reflected for a month straight. This is the #1 predictor of eventual success.
 
 ### The Override
 
@@ -274,7 +277,7 @@ But if you're overriding > 50% of the time for a month, ask yourself: _Is the sy
 
 ### For Consistency
 
-- **Run `loopkit loop` every Sunday at the same time.** Make it a ritual.
+- **Run `loopkit loop` every Sunday at the same time.** Make it a ritual. LoopKit will even provide a cron reminder hint after you initialize a project.
 - **Set a calendar reminder.** "Sunday 9 AM — LoopKit Loop"
 - **Share your BIP post every week.** Public commitment = accountability.
 

packages/cli/src/ai/prompts/init.ts

@@ -55,7 +55,7 @@ MVP SCOPE:
 - If MVP has more than 1 core feature, tell them to cut it
 - The validate action must be something they can do alone tonight — no "talk to 10 customers"
 - If Q2 answer is a solution not a problem, flag it: "That reads as a solution. What's the pain before your product exists?"
-- Max 250 words total in your entire output
+- Be concise: each field is 1-2 sentences max. Prefer precision over length.
 
 ## few-shot calibration
 

packages/cli/src/ai/prompts/loop.ts

@@ -64,26 +64,35 @@ export function buildLoopPrompt(context: {
   tasksOpen: string[];
   shipLog?: string;
   pulseData?: string;
+  previousWeeks?: Array<{ weekNumber: number; score: number; oneThing: string }>;
 }): string {
   const parts: string[] = [];
   parts.push(`Product: ${context.productName}`);
   parts.push(`Week: ${context.weekNumber}`);
   if (context.bet) parts.push(`Bet: ${context.bet}`);
   if (context.riskiestAssumption) parts.push(`Riskiest assumption: ${context.riskiestAssumption}`);
-  
+
+  if (context.previousWeeks && context.previousWeeks.length > 0) {
+    parts.push(
+      `\nRecent weeks (for accountability):\n${context.previousWeeks
+        .map((w) => `- Week ${w.weekNumber}: ${w.score}% score · one thing was "${w.oneThing}"`)
+        .join("\n")}`
+    );
+  }
+
   if (context.tasksCompleted.length > 0) {
     parts.push(`\nTasks completed:\n${context.tasksCompleted.map((t) => `- ${t}`).join("\n")}`);
   } else {
     parts.push("\nTasks completed: None this week.");
   }
-  
+
   if (context.tasksOpen.length > 0) {
     parts.push(`\nTasks still open:\n${context.tasksOpen.map((t) => `- ${t}`).join("\n")}`);
   }
-  
+
   if (context.shipLog) parts.push(`\nShip log:\n${context.shipLog}`);
   if (context.pulseData) parts.push(`\nPulse feedback:\n${context.pulseData}`);
-  
+
   parts.push("\nSynthesize this week and produce weekWin, oneThing, rationale, tension, bipPost, and founderNote.");
   return parts.join("\n");
 }

packages/cli/src/commands/celebrate.ts

@@ -210,6 +210,19 @@ export async function celebrateCommand(
   console.log(colors.dim("  Copy this to share your progress:"));
   console.log(box(shareText));
 
+  // Auto-copy to clipboard on macOS
+  try {
+    const { spawnSync } = await import("node:child_process");
+    if (process.platform === "darwin") {
+      const result = spawnSync("pbcopy", [], { input: shareText, encoding: "utf-8" });
+      if (!result.error) {
+        console.log(pass("Copied to clipboard — paste and share!"));
+      }
+    }
+  } catch {
+    // Silent — clipboard is a convenience, not required
+  }
+
   // ─── What's next ──────────────────────────────────────────────
   console.log(header("What's Next"));
   console.log(info("Run `loopkit loop` to synthesize your week"));

packages/cli/src/commands/init.ts

@@ -361,6 +361,13 @@ export async function initCommand(
   }
 
   console.log(nextStep("track"));
+  console.log(
+    colors.dim(
+      "\n  💡 Set a Sunday reminder to close your week:\n" +
+      "     macOS: Add a weekly calendar event for Sunday 6pm\n" +
+      `     Or cron: 0 18 * * 0 cd ${process.cwd()} && loopkit loop\n`
+    )
+  );
   p.outro(colors.muted("Brief saved. Now build against it."));
 }
 

packages/cli/src/commands/loop.ts

@@ -92,6 +92,16 @@ export async function loopCommand(): Promise<void> {
   const tasksContent = readTasksFile(slug);
   const shipLog = readShipLog();
 
+  // ─── Build pulse context for AI ──────────────────────────────
+  const pulseResponses = readPulseResponses();
+  const pulseData =
+    pulseResponses.length > 0
+      ? `${pulseResponses.length} responses. Recent: ${pulseResponses
+          .slice(-5)
+          .map((r, i) => `${i + 1}. "${r}"`)
+          .join(" | ")}`
+      : undefined;
+
   // Parse tasks
   let tasksCompleted: string[] = [];
   let tasksOpen: string[] = [];
@@ -114,6 +124,24 @@ export async function loopCommand(): Promise<void> {
       ? Math.round((tasksCompleted.length / totalTasks) * 100)
       : 0;
 
+  // ─── Build previousWeeks for AI accountability context ───────
+  const previousLogsRaw = readLastNLoopLogs(3, slug).filter(
+    (l) => l.weekNumber !== weekNum,
+  );
+  const previousWeeks = previousLogsRaw
+    .map((l) => {
+      const content = readLoopLog(l.weekNumber);
+      if (!content) return null;
+      const scoreMatch = content.match(/[Ss]hipping score:\s*(\d+)%/);
+      const oneThingMatch = content.match(/## The One Thing\n([^\n]+)/);
+      return {
+        weekNumber: l.weekNumber,
+        score: scoreMatch ? parseInt(scoreMatch[1], 10) : 0,
+        oneThing: oneThingMatch ? oneThingMatch[1].trim() : "",
+      };
+    })
+    .filter(Boolean) as Array<{ weekNumber: number; score: number; oneThing: string }>;
+
   const proof = computeLoopProof({
     slug,
     weekNum,
@@ -148,6 +176,8 @@ export async function loopCommand(): Promise<void> {
   const currentStreak = pastStreak + 1;
   if (currentStreak >= 2) {
     console.log(`  ${colors.primary.bold("Streak:")} 🔥 ${currentStreak} consecutive weeks`);
+  } else if (pastStreak === 0 && proof.weeksActive >= 2) {
+    console.log(warn("  Streak reset — you missed a week. Let's start a new one today."));
   }
 
   // ─── Handle no data (first week) ─────────────────────────────
@@ -209,8 +239,8 @@ export async function loopCommand(): Promise<void> {
     return;
   }
 
-  // ─── Unstuck Mode: 0 tasks this week ──────────────────────────
-  if (totalTasks === 0 && tasksCompleted.length === 0) {
+  // ─── Unstuck Mode: 0 tasks (only triggers after week 1) ──────
+  if (totalTasks === 0 && tasksCompleted.length === 0 && proof.weeksActive >= 2) {
     console.log(colors.warning("\n  No tasks this week. Feeling stuck?\n"));
 
     const wantUnstuck = await p.confirm({
@@ -302,6 +332,8 @@ export async function loopCommand(): Promise<void> {
         tasksCompleted,
         tasksOpen,
         shipLog: shipLog || undefined,
+        pulseData,
+        previousWeeks: previousWeeks.length > 0 ? previousWeeks : undefined,
       }),
       schema: LoopSynthesisSchema,
       tier: "fast",
@@ -612,11 +644,14 @@ function formatScoreDelta(delta: number): string {
 }
 
 async function maybeShowUpgradeIntent(proof: LoopProof): Promise<void> {
-  if (proof.weeksActive !== 1) return;
+  if (proof.weeksActive !== 4 && proof.weeksActive !== 8) return;
 
-  const wantsUpgrade = await p.confirm({
-    message: "Want the dashboard, hosted Pulse, and AI proxy when this loop starts compounding?",
-  });
+  const isWeek8 = proof.weeksActive === 8;
+  const message = isWeek8
+    ? `${proof.weeksActive} weeks of data — want the dashboard and AI proxy to go deeper?`
+    : "You're 4 weeks in — this is when the dashboard starts compounding. Want to unlock it?";
+
+  const wantsUpgrade = await p.confirm({ message });
 
   if (p.isCancel(wantsUpgrade) || !wantsUpgrade) return;
 

packages/cli/src/commands/pulse.ts

@@ -106,9 +106,7 @@ export async function pulseCommand(options: PulseOptions): Promise<void> {
 
       console.log(
         colors.muted(
-          '\n  Embed widget: <script src="' +
-            API_URL +
-            '/api/pulse/widget?projectId=..."></script>\n',
+          `\n  Embed widget: <script src="${API_URL}/api/pulse/widget?projectId=${convexProjectId || 'YOUR_PROJECT_ID'}"></script>\n`,
         ),
       );
     } catch (err) {
@@ -124,13 +122,25 @@ export async function pulseCommand(options: PulseOptions): Promise<void> {
     return;
   }
 
-  // ─── --add: Append a single response ─────────────────────────
+  // ─── --add: Append a single response ────────────────────────────────────
   if (options.add !== undefined) {
     const text = options.add.trim();
     if (!text) {
       console.log(colors.danger("Response text cannot be empty."));
       process.exit(1);
     }
+
+    // Dedup: skip exact duplicates (normalized)
+    const existing = readPulseResponses();
+    const normalized = text.toLowerCase().replace(/\s+/g, " ");
+    const isDuplicate = existing.some(
+      (r) => r.toLowerCase().replace(/\s+/g, " ") === normalized
+    );
+    if (isDuplicate) {
+      console.log(colors.warning("Response already exists (skipped duplicate)."));
+      return;
+    }
+
     appendPulseResponse(text);
     const total = readPulseResponses().length;
     console.log(pass(`Response added (${total} total)`));