crisis-detector
À propos
Cette compétence aide les développeurs à identifier les signaux d'alerte précoces de crises potentielles liées aux relations publiques grâce à la reconnaissance de motifs et à l'évaluation des risques. Elle est utile pour mettre en place des systèmes d'alerte précoce, évaluer la probabilité d'une crise et établir des critères d'escalade. L'outil s'appuie sur des cadres de gestion de crise éprouvés pour surveiller les signaux et évaluer les déclencheurs d'escalade.
Installation rapide
Claude Code
Recommandénpx skills add guia-matthieu/clawfu-skills -a claude-code/plugin add https://github.com/guia-matthieu/clawfu-skillsgit clone https://github.com/guia-matthieu/clawfu-skills.git ~/.claude/skills/crisis-detectorCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Crisis Detector
Identify early warning signs of potential crises before they escalate through pattern recognition, signal monitoring, and risk assessment.
When to Use This Skill
- Setting up early warning systems
- Assessing crisis probability
- Training teams on signals
- Building escalation criteria
- Post-crisis prevention planning
Methodology Foundation
Based on Institute for Crisis Management research and Burson crisis frameworks, combining:
- Signal identification
- Pattern recognition
- Risk assessment matrices
- Escalation protocols
What Claude Does vs What You Decide
| Claude Does | You Decide |
|---|---|
| Identifies warning signals | Risk tolerance |
| Assesses crisis probability | Response resources |
| Creates detection criteria | Escalation authority |
| Designs monitoring systems | Communication strategy |
| Suggests response triggers | Final action calls |
Instructions
Step 1: Map Crisis Types
Crisis Categories:
| Category | Examples | Warning Time |
|---|---|---|
| Operational | Outage, product failure | Hours to days |
| Reputational | Executive scandal, viral complaint | Minutes to hours |
| Legal/Regulatory | Lawsuit, investigation | Days to weeks |
| Financial | Earnings miss, fraud | Hours to days |
| Human | Workplace incident, harassment | Hours to days |
| External | Natural disaster, market crash | Variable |
Step 2: Identify Early Signals
Signal Types:
| Signal Type | Examples | Monitoring |
|---|---|---|
| Internal | Employee complaints, support tickets | HR, Support data |
| Customer | Review patterns, churn spikes | CX metrics |
| Social | Mention volume, sentiment shift | Social tools |
| Media | Press inquiries, journalist interest | PR inbox |
| Regulatory | Compliance notices, audit findings | Legal |
| Financial | Payment disputes, refund requests | Finance |
Step 3: Build Detection Matrix
Signal Strength Assessment:
| Signal | Weak | Moderate | Strong | Critical |
|---|---|---|---|---|
| Volume spike | +25% | +50% | +100% | +300% |
| Sentiment shift | -10% | -20% | -30% | -50% |
| Media inquiries | 1 | 2-3 | 4-5 | 6+ |
| Support tickets | +10% | +25% | +50% | +100% |
| Social influencer | 10K | 50K | 100K | 500K+ |
Step 4: Create Escalation Triggers
Trigger Framework:
| Level | Signals Present | Action |
|---|---|---|
| Watch | 1 moderate signal | Monitor closely |
| Alert | 2+ moderate or 1 strong | Notify team |
| Warning | Multiple strong signals | Assemble team |
| Crisis | Any critical signal | Activate protocol |
Step 5: Document and Improve
Post-Detection Review:
- How early did we detect?
- What signals did we miss?
- How accurate was assessment?
- What should we add to monitoring?
Examples
Example 1: Crisis Detection System Design
Input:
Design a crisis detection system for:
- E-commerce company (clothing)
- 500K customers
- Active on Instagram, TikTok, Twitter
- Past issue: Shipping delays during holiday
Output:
## Crisis Detection System: E-Commerce Fashion
### Crisis Risk Map
| Crisis Type | Probability | Impact | Priority |
|-------------|-------------|--------|----------|
| Shipping delays | High | High | P1 |
| Product quality issue | Medium | High | P1 |
| Social media backlash | Medium | High | P1 |
| Data breach | Low | Critical | P1 |
| Influencer controversy | Medium | Medium | P2 |
| Supply chain disruption | Medium | High | P2 |
| Payment fraud | Low | Medium | P3 |
---
### Early Warning Signals
#### P1: Shipping Delays
**Leading Indicators (3-5 days before crisis):**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Carrier delay reports | Logistics API | >10% delayed |
| Warehouse backlog | WMS data | >24hr processing |
| Weather events | News/weather | Storm in hub |
| "Where's my order" tickets | Support | +50% daily |
**Lagging Indicators (crisis starting):**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Social mentions | Social listening | "shipping" +100% |
| Review mentions | Trustpilot/G2 | Shipping 3/5 stars |
| Refund requests | Payment system | +30% |
| Chargeback rate | Payment processor | >1% |
---
#### P1: Product Quality Issue
**Leading Indicators:**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Return rate spike | Returns data | >10% on SKU |
| Quality complaints | Support tickets | 3+ same issue |
| Photo complaints | Social | "damaged", "wrong color" |
| Batch-specific issues | QC data | Same lot number |
**Lagging Indicators:**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Viral unboxing | TikTok/Instagram | >10K views negative |
| Review bomb | Product pages | Multiple 1-stars |
| Media inquiry | PR inbox | Journalist question |
---
#### P1: Social Media Backlash
**Leading Indicators:**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Sentiment shift | Social tools | -20% in 24hr |
| Controversial post | Your social | Negative comments >10% |
| Influencer complaint | Social | >50K follower post |
| Screenshot spreading | Twitter/Reddit | Same image 5+ times |
**Lagging Indicators:**
| Signal | Source | Threshold |
|--------|--------|-----------|
| Viral negative | Any platform | >50K engagements |
| Hashtag trending | Twitter | Brand + negative |
| Media pickup | News sites | Article published |
| Competitor amplification | Social | Competitor sharing |
---
### Detection Dashboard
┌──────────────────────────────────────────────────────────┐ │ CRISIS DETECTION DASHBOARD 🟢 NORMAL │ ├──────────────────────────────────────────────────────────┤ │ │ │ SHIPPING STATUS 🟢 Normal │ │ ├─ Carrier delays: 3% (threshold: 10%) │ │ ├─ Backlog: 4 hours (threshold: 24hr) │ │ └─ "Where's my order": 45 (baseline: 50) │ │ │ │ PRODUCT QUALITY 🟢 Normal │ │ ├─ Return rate: 5.2% (threshold: 10%) │ │ ├─ Quality tickets: 2 (threshold: 3+ same) │ │ └─ Photo complaints: 1 (threshold: 5) │ │ │ │ SOCIAL SENTIMENT 🟡 Watch │ │ ├─ Sentiment: -15% (threshold: -20%) │ │ ├─ Volume: +40% (threshold: +50%) │ │ └─ Influencer: None flagged │ │ │ │ SECURITY 🟢 Normal │ │ ├─ Login anomalies: Normal │ │ └─ Payment fraud: 0.3% │ │ │ └──────────────────────────────────────────────────────────┘
---
### Escalation Matrix
| Level | Criteria | Notification | Response |
|-------|----------|--------------|----------|
| 🟢 Normal | All metrics in range | None | Standard ops |
| 🟡 Watch | 1 metric at threshold | Slack alert | Increased monitoring |
| 🟠 Alert | 2+ metrics or 1 exceeded | Team notification | Standby team |
| 🔴 Crisis | Critical threshold breached | All-hands alert | Activate protocol |
---
### Alert Routing
| Signal Type | Primary | Backup | Escalation |
|-------------|---------|--------|------------|
| Shipping | Operations | COO | CEO |
| Product quality | Product | VP Ops | CEO |
| Social media | Marketing | CMO | CEO |
| Security | IT | CTO | CEO + Legal |
| Legal/PR | Legal | CEO | Board |
---
### Holiday Season Adjustments
During high-volume periods (Nov 15 - Dec 31):
| Metric | Normal Threshold | Holiday Threshold |
|--------|------------------|-------------------|
| Carrier delays | 10% | 15% |
| Support volume | +50% | +100% |
| Social volume | +50% | +75% |
| Response time | 4 hours | 8 hours |
**Rationale:** Higher baseline expected; adjust to avoid false alerts
while maintaining genuine crisis detection.
Example 2: Real-Time Crisis Assessment
Input:
Assess this situation:
- Support tickets up 80% in 2 hours
- Twitter mentions up 150%
- Hashtag #[Brand]Scam trending locally
- 3 refund requests mentioning "charged twice"
- No known system issues
Output:
## Crisis Assessment: Potential Payment Issue
### Alert Level: 🔴 CRISIS
---
### Signal Analysis
| Signal | Value | Normal | Status |
|--------|-------|--------|--------|
| Support spike | +80% | <25% | CRITICAL |
| Social spike | +150% | <50% | CRITICAL |
| Trending hashtag | Yes | No | CRITICAL |
| Duplicate charge reports | 3 | 0 | CONCERNING |
---
### Probable Cause
**Most likely: Payment processing error**
Evidence:
1. Multiple "charged twice" complaints
2. No known system issues rules out outage
3. Sudden spike suggests batch problem
4. #Scam hashtag = customers think fraud
---
### Immediate Actions
| Priority | Action | Owner | Timeline |
|----------|--------|-------|----------|
| 1 | Check payment processor logs | Engineering | NOW |
| 2 | Identify affected transactions | Finance | 30 min |
| 3 | Prepare holding statement | Comms | 15 min |
| 4 | Alert customer service team | CX Lead | NOW |
| 5 | Monitor hashtag spread | Social | Ongoing |
---
### Holding Statement (Draft)
We're aware some customers may have experienced duplicate charges. Our team is investigating urgently. If you've been affected, please DM us or email [support] - we'll make this right immediately.
Updates to follow shortly.
---
### Escalation Path
**Now:** CTO + CFO + CMO notified
**+30 min:** CEO briefing if not resolved
**+1 hour:** External statement if ongoing
---
### Crisis Trajectory
**If unaddressed (next 2-4 hours):**
- Hashtag goes national
- Media inquiries begin
- Trust pilot review bomb
- Social influencers amplify
**If addressed quickly (next 1-2 hours):**
- Contain to affected customers
- Flip narrative to "responsive company"
- Prevent media escalation
- Build goodwill through fast resolution
---
### Resolution Checklist
- [ ] Root cause identified
- [ ] Affected customers identified
- [ ] Refunds initiated
- [ ] Proactive communication sent
- [ ] Social response deployed
- [ ] Hashtag monitoring active
- [ ] Post-incident review scheduled
Skill Boundaries
What This Skill Does Well
- Identifying early warning signals
- Creating detection frameworks
- Assessing crisis probability
- Designing escalation systems
What This Skill Cannot Do
- Access your actual systems
- Monitor in real-time
- Make response decisions
- Know your specific thresholds
Iteration Guide
Follow-up Prompts:
- "Design detection for [specific crisis type]"
- "Create escalation protocol for [scenario]"
- "What signals should we add for [risk]?"
- "How do we prevent [past crisis] from recurring?"
References
- Institute for Crisis Management
- Burson Crisis Playbook
- Harvard Business Review Crisis Research
- Edelman Trust Barometer
Related Skills
social-listening- Monitoring systemsresponse-coordinator- Crisis responsereputation-recovery- Post-crisis rebuild
Skill Metadata
- Domain: Crisis
- Complexity: Intermediate-Advanced
- Mode: centaur
- Time to Value: 2-4 hours for system design
- Prerequisites: Access to metrics, stakeholder alignment
Dépôt GitHub
Compétences associées
llamaguard
AutreLlamaGuard est le modèle de Meta, doté de 7 à 8 milliards de paramètres, conçu pour modérer les entrées et sorties des LLM selon six catégories de sécurité comme la violence et les discours haineux. Il offre une précision de 94 à 95 % et peut être déployé avec vLLM, Hugging Face ou Amazon SageMaker. Utilisez cette compétence pour intégrer facilement le filtrage de contenu et des garde-fous de sécurité dans vos applications d'IA.
cost-optimization
AutreCette compétence de Claude aide les développeurs à optimiser les coûts du cloud grâce au redimensionnement des ressources, aux stratégies d'étiquetage et à l'analyse des dépenses. Elle fournit un cadre pour réduire les dépenses cloud et mettre en œuvre une gouvernance des coûts sur AWS, Azure et GCP. Utilisez-la lorsque vous devez analyser les coûts d'infrastructure, redimensionner les ressources ou respecter des contraintes budgétaires.
quantizing-models-bitsandbytes
AutreCette compétence quantifie les LLMs en précision 8 bits ou 4 bits à l'aide de bitsandbytes, permettant une réduction de 50 à 75 % de la mémoire utilisée avec une perte de précision minime. Elle est idéale pour exécuter des modèles plus volumineux sur une mémoire GPU limitée ou pour accélérer l'inférence, prenant en charge des formats comme INT8, NF4 et FP4. La compétence s'intègre à HuggingFace Transformers et permet l'entraînement QLoRA ainsi que l'utilisation d'optimiseurs en 8 bits.
dispatching-parallel-agents
AutreCette compétence Claude déploie plusieurs agents pour enquêter et résoudre simultanément 3 problèmes indépendants ou plus. Elle est conçue pour des scénarios impliquant des défaillances non liées qui peuvent être résolues sans état partagé ni dépendances. La capacité fondamentale est la résolution de problèmes en parallèle, en assignant un agent par domaine problématique indépendant afin de maximiser l'efficacité.
