@seanyao/roll 3.609.2 → 3.610.2

package/skills/roll-sentinel/SKILL.md

@@ -2,363 +2,45 @@
 name: roll-sentinel
 license: MIT
 allowed-tools: "Read, Edit, Write, Bash, WebFetch"
-description: Smart patrol inspector for production systems. Scheduled randomized sampling checks based on BACKLOG requirements. Cost-controlled AI validation with intelligent spot-checking logic.
+description: "Load when production or deployed behavior needs cost-controlled randomized patrol checks based on backlog requirements and runtime sampling."
 ---
+# Roll Sentinel
 
-# Sentinel
+This hub keeps the routing boundary, hard gates, and execution skeleton in the initial context. Load the heavier runbook only when the task actually needs the detailed contract.
 
-**Smart Patrol Inspector** - Scheduled, randomized, cost-controlled patrol and acceptance checks for production systems.
+## Load
 
-## When Not to Use
-
-- One-off debugging of a reported bug (use `$roll-debug`)
-- Full-coverage regression testing (sentinel samples, does not cover)
-- Dev/staging environment checks (use CI tests instead)
-- Pre-commit self-review of diffs (use `$roll-.review`)
-
-## Core Principle
-
-```
-┌─────────────────────────────────────────────────────────────┐
-│                   Smart Patrol Logic                        │
-├─────────────────────────────────────────────────────────────┤
-│                                                             │
-│  Not full-coverage checks! Think of it like a security      │
-│  guard on patrol:                                           │
-│                                                             │
-│  🕐 Scheduled Triggers - Auto-patrol on schedule            │
-│       └── "Patrol once every 6 hours"                       │
-│                                                             │
-│  🎲 Random Sampling - Different samples each time           │
-│       └── "Check Stories 1-10 this time, 50-60 next time"   │
-│                                                             │
-│  💰 Cost Control - AI checks are expensive, use sparingly   │
-│       └── "Only check 10 each time, not all 100"            │
-│                                                             │
-│  🎯 BACKLOG-Based - Validate against requirements           │
-│       └── "US-001 says login works, so verify login"        │
-│                                                             │
-└─────────────────────────────────────────────────────────────┘
-```
-
-## Patrol Strategy
-
-### Sampling Logic
-
-```javascript
-// Sampling logic for each patrol
-function selectSamples(backlog, strategy = 'smart') {
-  const completedStories = backlog.filter(s => s.status === '✅');
-  
-  switch(strategy) {
-    case 'random':
-      // Fully random: randomly select N from all completed Stories
-      return shuffle(completedStories).slice(0, 10);
-    
-    case 'weighted':
-      // Weighted random: prioritize recently modified and frequently used
-      return completedStories
-        .sort((a, b) => b.lastModified - a.lastModified)
-        .slice(0, 5)  // 5 most recent
-        .concat(shuffle(completedStories).slice(0, 5)); // + 5 random
-    
-    case 'coverage':
-      // Coverage sampling: ensure different modules are all covered
-      const byModule = groupBy(completedStories, 'module');
-      return Object.values(byModule).map(
-        stories => randomPick(stories)
-      );
-  }
-}
-```
-
-### Cost Control
-
-| Strategy | Sample Size | Frequency | Use Case |
-|----------|-------------|-----------|----------|
-| **Light** | 5 Stories | Once daily | Stable period |
-| **Normal** | 10 Stories | Every 6 hours | General monitoring |
-| **Intensive** | 20 Stories | Every hour | Post-release period |
-| **Full** | All | Once weekly | Weekly patrol |
-
-```yaml
-# sentinel.config.yml
-cost_control:
-  daily_budget: 100  # AI call budget
-  
-  light_patrol:
-    samples: 5
-    schedule: "0 9 * * *"  # Daily at 9am
-    
-  normal_patrol:
-    samples: 10
-    schedule: "0 */6 * * *"  # Every 6 hours
-    
-  # Intensive patrol after deployment
-  post_deploy:
-    trigger: "after_deploy"
-    samples: 20
-    duration: "2h"  # Lasts 2 hours
-```
-
-### Uncertainty Handling
-
-```javascript
-// Systems have uncertainty; a single check may be inaccurate.
-// Use multiple random checks to increase confidence.
-
-class UncertaintyHandler {
-  // Track check result history
-  history = new Map(); // storyId -> [check1, check2, ...]
-  
-  // Determine if an issue is real
-  isRealIssue(storyId, currentResult) {
-    const pastResults = this.history.get(storyId) || [];
-    pastResults.push(currentResult);
-    
-    // Only consider it a real issue if it fails 3 times consecutively
-    const recent3 = pastResults.slice(-3);
-    if (recent3.every(r => r.status === 'FAIL')) {
-      return true; // Real issue
-    }
-    
-    // If it fails occasionally, it may be intermittent; keep observing
-    if (recent3.filter(r => r.status === 'FAIL').length === 1) {
-      return false; // Likely intermittent, don't alert yet
-    }
-    
-    return false;
-  }
-}
-```
-
-## When to Patrol
-
-### Scheduled Patrols
-
-```bash
-# Daily patrol - randomly check a few each day
-$roll-sentinel patrol --mode=normal
-
-# Late-night patrol - full check during off-peak hours
-$roll-sentinel patrol --mode=full --schedule="0 3 * * *"
-
-# Weekend walkthrough - check the week's accumulation on Sunday
-$roll-sentinel patrol --mode=weekly --schedule="0 10 * * 0"
-```
-
-### Event-Triggered
-
-```bash
-# Intensive patrol for 2 hours after deployment
-$roll-sentinel patrol --mode=intensive --duration=2h --after-deploy
-
-# Emergency check after an alert
-$roll-sentinel patrol --mode=focus --target=US-XXX
-```
+Load when production or deployed behavior needs cost-controlled randomized patrol checks based on backlog requirements and runtime sampling.
 
-## Patrol Report
-
-```markdown
-## 🛡️ Sentinel Patrol Report #247
-**Time**: 2024-01-15 14:00 UTC  
-**Patrol ID**: patrol-20240115-1400  
-**Mode**: Normal (Random Sampling)
-
-### 📊 Sampling Info
-| Metric | Value |
-|--------|-------|
-| Total Stories | 150 |
-| Sample Size | 10 |
-| Sampling Rate | 6.7% |
-| Cost Estimate | $0.07 |
-
-### 🎲 Random Sample
-| # | Story | Module | Last Checked | Result |
-|---|-------|--------|--------------|--------|
-| 1 | US-LOGIN-001 | Auth | 6h ago | ✅ |
-| 2 | US-STORY-042 | Content | 12h ago | ✅ |
-| 3 | US-AUDIO-015 | Player | 2h ago | 🟡* |
-| 4 | US-SEARCH-003 | Search | 18h ago | ✅ |
-| 5 | ... | ... | ... | ... |
-
-\* US-AUDIO-015: Occasional playback stuttering (2nd occurrence, under observation)
-
-### 🔴 Issues Found
-None (no confirmed issues found in this sample)
-
-### 📈 Patrol Statistics (7 days)
-| Metric | Value |
-|--------|-------|
-| Total Patrols | 28 |
-| Stories Checked | 280 |
-| Issues Found | 3 |
-| False Positives | 1 |
-| Coverage | 85% of all stories |
-
-### 💰 Cost Report
-| Item | Usage |
-|------|-------|
-| AI Checks | 280 calls |
-| Playwright Runs | 28 sessions |
-| Total Cost | $2 |
-| Budget Used | 15% of monthly |
-```
-
-## Smart Detection Logic
-
-### Pattern 1: Intermittent vs Real Issue
-
-```javascript
-// Don't alert on the first failure; look at the trend
-const checks = [
-  { time: 'T-6h', status: 'PASS' },
-  { time: 'T-12h', status: 'FAIL' },  // Intermittent?
-  { time: 'T-18h', status: 'PASS' },
-  { time: 'Now', status: 'FAIL' },    // Failed again!
-];
-
-// 2 consecutive failures → Create Issue
-if (lastN(checks, 2).all(c => c.status === 'FAIL')) {
-  createBacklogItem('FIX-XXX');
-}
-// Occasional failure → Log for observation
-else if (checks.filter(c => c.status === 'FAIL').length <= 1) {
-  logForObservation('Might be flaky, continue monitoring');
-}
-```
-
-### Pattern 2: Hotspot Detection
-
-```javascript
-// Some Stories frequently show issues when sampled.
-// Automatically increase their check frequency.
-
-const hotSpots = analyzeHistory();
-// hotSpots = [
-//   { story: 'US-AUDIO-015', failRate: 0.3 }, // 30% failure rate
-//   { story: 'US-SEARCH-003', failRate: 0.1 },
-// ]
-
-// Increase weight for hotspots
-if (hotSpots.some(h => h.story === selectedStory)) {
-  // If it's a hotspot, even if not randomly selected, add extra check probability
-  if (Math.random() < 0.3) {
-    extraCheck(story);
-  }
-}
-```
-
-## Cost Optimization
-
-### Tiered Checking
-
-```
-Level 1: Lightweight Check (cheap)
-  └── HTTP ping / API health check
-  └── Cost: $0.001 per check
-
-Level 2: Functional Check (moderate)
-  └── Playwright critical path
-  └── Cost: $0.01 per check
+## When Not to Use
 
-Level 3: AI Deep Check (expensive)
-  └── AI-powered content quality analysis
-  └── Cost: $0.07 per check
+- Code health scans; load roll-.dream.
+- Interactive page debugging; load roll-debug.
 
-Strategy:
-- Each patrol: 90% Level 1 + 10% Level 2
-- Once weekly: Level 3 deep inspection
-```
+## Read On Demand
 
-### Smart Batching
+- Read [the full contract](references/full-contract.md) before executing the workflow end to end, recovering from failures, or checking exact output templates.
+- Keep this hub in context for trigger boundaries and hard gates.
 
-```javascript
-// Batch checks to reduce cost.
-// Instead of 10 separate checks, open one browser for all 10.
+## Workflow Skeleton
 
-async function batchCheck(stories) {
-  const browser = await chromium.launch();
-  const context = await browser.newContext();
-  
-  // Reuse browser session to check multiple Stories
-  for (const story of stories) {
-    const page = await context.newPage();
-    await checkStory(page, story);
-    await page.close();
-  }
-  
-  await browser.close();
-  // Cost: 1 session for 10 checks
-}
-```
+1. Select sampling targets from backlog requirements.
+2. Run randomized cost-controlled checks.
+3. Validate observed production behavior.
+4. Record patrol evidence and escalate anomalies.
 
-## Workflow: Find Issue → Backlog
+## Hard Gates
 
-```
-┌─────────────────────────────────────────────────────────────┐
-│              Issue Discovery Workflow via Patrol             │
-├─────────────────────────────────────────────────────────────┤
-│                                                             │
-│  1. Sentinel Patrol (scheduled random sampling)             │
-│     └── Sample: US-AUDIO-015                                │
-│                                                             │
-│  2. Check Result                                            │
-│     └── Status: FAIL (playback stuttering)                  │
-│                                                             │
-│  3. Uncertainty Check                                       │
-│     └── Check history: this is the 2nd failure              │
-│     └── 1st was 6h ago (possibly intermittent)              │
-│     └── Decision: continue observing, don't create Issue    │
-│                                                             │
-│  4. Next Patrol                                             │
-│     └── US-AUDIO-015 sampled again (hotspot weighting)      │
-│     └── Status: FAIL (failed again!)                        │
-│     └── Check history: 2 consecutive failures               │
-│     └── Decision: create FIX-AUDIO-015                      │
-│                                                             │
-│  5. Create Backlog Item                                     │
-│     └── Add FIX-AUDIO-015 to .roll/backlog.md                     │
-│     └── Status: 📋 Todo                                     │
-│     └── Awaiting human fix                                  │
-│                                                             │
-│  6. Human Fix                                               │
-│     └── User: "Fix FIX-AUDIO-015"                           │
-│     └── $roll-fix FIX-AUDIO-015                            │
-│                                                             │
-│  7. Verification                                            │
-│     └── Next patrol will prioritize verifying this FIX      │
-│     └── Status: ✅ Fixed                                     │
-│                                                             │
-└─────────────────────────────────────────────────────────────┘
-```
+- Sampling budget remains explicit.
+- Do not turn sentinel into exhaustive monitoring.
 
-## Integration with Other Skills
+## Gotchas
 
-```
-┌─────────────────────────────────────────────────────────────┐
-│                Complete Monitoring System                    │
-├─────────────────────────────────────────────────────────────┤
-│                                                             │
-│  $roll-sentinel patrol        Scheduled random patrol (main) │
-│       ↓                                                     │
-│  Issue found? ──┬── Yes ──→ Create BACKLOG item             │
-│                 │           Await $roll-fix             │
-│                 │                                           │
-│                 └── No  ──→ Continue patrolling              │
-│                                                             │
-│  $roll-debug               On-demand deep diagnosis (aux) │
-│  (When Sentinel finds an issue, manually trigger deep dive) │
-│                                                             │
-│  $roll-build            Post-fix regression verify     │
-│                                                             │
-└─────────────────────────────────────────────────────────────┘
-```
+- Sentinel samples deployed behavior; it is not a code architecture scanner.
+- Keep cost controls and randomized sampling visible; do not turn patrols into exhaustive monitoring.
 
-## Best Practices
+## Maintenance
 
-1. **Don't do full-coverage checks** - Expensive and unnecessary
-2. **Random + Hotspots** - Balance coverage and cost
-3. **Multi-check confirmation** - Avoid false positives from intermittent failures
-4. **Budget control** - Set daily/monthly AI call limits
-5. **Progressive intensity** - Light during stable periods, Intensive after releases
+- Description changes require updates in `route-cases/skills.json`.
+- New observed failures should add a gotcha and the matching positive or negative route case.
+- Heavy examples, templates, recovery paths, and deterministic snippets belong in `references/`, `assets/`, or `scripts/`, not in this hub.