Menú
De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.
Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Un comando directo omite el prompt de revisión. Revisa el origen antes de ejecutarlo.
npx skills add https://github.com/dvcrn/openclaw-skills-marketplace --skill afrexai-automation-strategyEl comando permanece en una sola línea. Desplázate horizontalmente para revisarlo antes de copiarlo.
¿Prefieres una copia local? Descarga los archivos que SkillsMP tiene disponibles ahora.
Basado en la clasificación ocupacional SOC
Mostrando SKILL.md
| name | afrexai-automation-strategy |
| description | Business Automation Strategy |
The complete methodology for identifying, designing, building, and scaling business automations. Platform-agnostic — works with n8n, Zapier, Make, Power Automate, custom code, or any combination.
Before building anything, map where time and money leak.
Ask these 5 questions about any process:
process_inventory:
process_name: "[Name]"
department: "[Sales/Marketing/Ops/Finance/HR/Engineering]"
owner: "[Person responsible]"
frequency: "[X per day/week/month]"
duration_minutes: [time per occurrence]
monthly_volume: [total occurrences]
monthly_hours: [volume × duration ÷ 60]
[ ]
[ ]
[, ]
[, ]
| Level | Description | Human Role | Example |
|---|---|---|---|
| Full | End-to-end automated, no human needed | Monitor exceptions | Invoice processing, data sync |
| Partial | Automated with human approval gates | Review & approve | Contract generation, hiring workflow |
| Assist | Human does work, automation helps | Execute with AI assistance | Customer support, content creation |
| None | Requires human judgment/creativity | Full ownership | Strategy, relationship building |
Annual savings = (monthly_hours × 12 × hourly_cost) + (error_rate × volume × 12 × error_cost)
Build cost = development_hours × developer_rate + tool_costs
Payback period = build_cost ÷ (annual_savings ÷ 12) months
ROI = ((annual_savings - annual_tool_cost) ÷ build_cost) × 100%
Decision rules:
| Dimension | Weight | Scoring Guide |
|---|---|---|
| Impact | 30% | 10=saves >$50K/yr, 7=saves >$20K/yr, 5=saves >$5K/yr, 3=saves >$1K/yr |
| Confidence | 20% | 10=proven pattern, 7=similar done before, 5=feasible but new, 3=uncertain |
| Ease | 25% | 10=<1 day, 7=<1 week, 5=<1 month, 3=<3 months, 1=>3 months |
| Reliability | 25% | 10=deterministic, 7=95%+ success, 5=80%+ success, 3=needs frequent fixes |
Score = (Impact × 0.30) + (Confidence × 0.20) + (Ease × 0.25) + (Reliability × 0.25)
Automate FIRST (highest ROI, lowest risk):
Automate LAST (complex, high risk):
| Factor | No-Code (Zapier/Make) | Low-Code (n8n/Power Automate) | Custom Code | AI Agent |
|---|---|---|---|---|
| Best for | Simple integrations | Complex workflows | Unique logic | Judgment calls |
| Build speed | Hours | Days | Weeks | Days-weeks |
| Maintenance | Low | Medium | High | Medium |
| Flexibility | Limited | High | Unlimited | High |
| Cost at scale | Expensive | Moderate | Cheap | Varies |
| Error handling | Basic | Good | Full control | Variable |
| Team skill needed | Business user | Technical BA | Developer | AI engineer |
| Vendor lock-in | High | Medium | None | Low-medium |
Is the process deterministic (same input → same output)?
├── YES: Does it involve >3 systems?
│ ├── YES: Does it need complex branching logic?
│ │ ├── YES → Low-code (n8n/Power Automate)
│ │ └── NO → No-code (Zapier/Make) if budget allows, else n8n
│ └── NO: Is it performance-critical?
│ ├── YES → Custom code
│ └── NO → No-code (simplest wins)
└── NO: Does it need judgment/reasoning?
├── YES: Is the judgment pattern learnable?
│ ├── YES → AI agent with human review
│ └── NO → Human-assisted automation
└── NO → Partial automation with human gates
| Monthly Tasks | Zapier | Make | n8n (self-hosted) | Custom Code |
|---|---|---|---|---|
| 1,000 | $30 | $10 | $5 (hosting) | $50+ (hosting) |
| 10,000 | $100 | $30 | $5 | $50+ |
| 100,000 | $500+ | $150 | $10 | $50+ |
| 1,000,000 | $2,000+ | $500+ | $20 | $100+ |
Rule: If you're spending >$200/mo on Zapier/Make, evaluate self-hosted n8n.
workflow_blueprint:
name: "[Descriptive name]"
id: "WF-[DEPT]-[NUMBER]"
version: "1.0.0"
owner: "[Person]"
priority: "[P0-P3]"
trigger:
type: "[webhook/schedule/event/manual/condition]"
source: "[System or schedule]"
conditions: "[When to fire]"
dedup_strategy: "[How to prevent double-processing]"
inputs:
- name: "[field]"
type: "[string/number/date/object]"
required: true
validation: "[rules]"
source: "[where it comes from]"
steps:
- id: "step_1"
action: "[verb: fetch/transform/validate/send/create/update/delete]"
system: "[target system]"
description: "[what this step does]"
input: "[from trigger or previous step]"
output: "[what it produces]"
error_handling: "[retry/skip/alert/abort]"
timeout_seconds: 30
- id: "step_2_branch"
type: "condition"
condition: "[expression]"
true_path: "step_3a"
false_path: "step_3b"
error_handling:
retry_policy:
max_attempts: 3
backoff: "exponential"
initial_delay_seconds: 5
on_failure: "[alert/queue-for-review/fallback]"
alert_channel: "[Slack/email/SMS]"
dead_letter_queue: true
monitoring:
success_metric: "[what defines success]"
expected_duration_seconds: [max]
alert_on_duration_exceeded: true
log_level: "[info/debug/error]"
testing:
test_data: "[how to generate test inputs]"
expected_output: "[what success looks like]"
edge_cases: ["empty input", "duplicate", "malformed data"]
| Pattern | When to Use | Example |
|---|---|---|
| Sequential | Steps depend on each other | Lead → Enrich → Score → Route |
| Parallel fan-out | Independent steps | Send email + Update CRM + Log analytics |
| Conditional branch | Different paths by data | High value → Sales, Low value → Nurture |
| Loop/batch | Process collections | For each row in CSV, create record |
| Approval gate | Human judgment needed | Contract review before sending |
| Event-driven chain | Workflow triggers workflow | Order placed → Fulfillment → Shipping → Notification |
| Retry with fallback | Unreliable external APIs | Try API → Retry 3x → Use cached data → Alert |
| Scheduled sweep | Periodic cleanup/sync | Nightly: sync CRM → accounting |
For every system integration:
data_mapping:
source_system: "[System A]"
target_system: "[System B]"
sync_direction: "[one-way/bidirectional]"
sync_frequency: "[real-time/5min/hourly/daily]"
conflict_resolution: "[source wins/target wins/newest wins/manual]"
field_mappings:
- source_field: "contact.email"
target_field: "customer.email_address"
transform: "lowercase"
required: true
- source_field: "contact.company"
target_field: "customer.organization"
transform: "trim"
default: "Unknown"
- source_field: "contact.created_at"
target_field: "customer.signup_date"
transform: "ISO8601 → YYYY-MM-DD"
| Approach | When | Implementation |
|---|---|---|
| Queue + throttle | Predictable volume | Process queue at 80% of rate limit |
| Exponential backoff | Burst traffic | Wait 1s, 2s, 4s, 8s on 429 errors |
| Batch API calls | High volume CRUD | Group 50-100 records per call |
| Cache responses | Repeated lookups | Cache for TTL matching data freshness needs |
| Off-peak scheduling | Non-urgent syncs | Run heavy syncs at 2-4 AM |
| Type | Example | Response | Priority |
|---|---|---|---|
| Transient | API timeout, 503 | Retry with backoff | Auto-handle |
| Rate limit | 429 Too Many Requests | Queue + throttle | Auto-handle |
| Data validation | Missing required field | Log + skip + alert | Review daily |
| Auth failure | Token expired | Refresh + retry, else alert | P1 — fix within 1h |
| Logic error | Unexpected state | Halt + alert + queue | P0 — fix immediately |
| External change | API schema changed | Halt + alert | P0 — fix immediately |
| Capacity | Queue overflow | Scale + alert | P1 — fix within 4h |
Every workflow should have a DLQ:
States: CLOSED (normal) → OPEN (failing) → HALF-OPEN (testing)
CLOSED: Process normally, track failures
→ If failure_count > threshold in window → OPEN
OPEN: Reject all requests, return cached/default
→ After cool_down_period → HALF-OPEN
HALF-OPEN: Allow 1 test request
→ If success → CLOSED
→ If failure → OPEN (reset cool_down)
Thresholds:
| Level | What | How | When |
|---|---|---|---|
| Unit | Individual step logic | Mock inputs, verify output | Every change |
| Integration | System connections | Test with sandbox APIs | Weekly + after changes |
| End-to-end | Full workflow path | Run with test data | Before deploy + weekly |
| Chaos | Failure scenarios | Kill steps, corrupt data | Monthly |
| Load | Volume handling | 10x normal volume | Before scaling |
For every workflow, test:
go_live_checklist:
functionality:
- [ ] All test scenarios pass
- [ ] Edge cases documented and handled
- [ ] Error messages are actionable
reliability:
- [ ] Retry logic tested
- [ ] Circuit breaker configured
- [ ] Dead letter queue active
- [ ] Idempotency verified (run twice, same result)
monitoring:
- [ ] Success/failure alerts configured
- [ ] Duration alerts set
- [ ] Log retention configured
- [ ] Dashboard created
documentation:
- [ ] Workflow blueprint updated
- [ ] Runbook written
- [ ] Team trained on manual override
rollback:
- [ ] Previous version preserved
- [ ] Rollback procedure tested
- [ ] Data cleanup plan for partial runs
automation_dashboard:
period: "weekly"
summary:
total_workflows: [count]
total_executions: [count]
success_rate: "[X%]"
avg_duration: "[X seconds]"
errors_this_period: [count]
time_saved_hours: [calculated]
cost_saved: "$[calculated]"
by_workflow:
- name: "[Workflow name]"
executions: [count]
success_rate: "[X%]"
avg_duration: "[X seconds]"
p95_duration: "[X seconds]"
errors: [count]
error_types: ["type1: count", "type2: count"]
dlq_items: [count]
status: "[healthy/degraded/failing]"
alerts_fired: [count]
manual_interventions: [count]
top_issues:
- "[Issue 1: description + fix status]"
- "[Issue 2: description + fix status]"
cost:
platform_cost: "$[monthly]"
api_calls_cost: "$[monthly]"
compute_cost: "$[monthly]"
total: "$[monthly]"
cost_per_execution: "$[calculated]"
| Metric | Warning | Critical | Action |
|---|---|---|---|
| Success rate | <95% | <90% | Investigate + fix |
| Duration | >2x average | >5x average | Check for bottleneck |
| DLQ size | >10 items | >50 items | Review + reprocess |
| Error spike | 5 errors/hour | 20 errors/hour | Pause + investigate |
| Queue depth | >100 pending | >1000 pending | Scale + investigate |
| Cost spike | >150% of average | >300% of average | Audit + optimize |
Before scaling any automation:
| Signal | From | To |
|---|---|---|
| Spending >$500/mo on Zapier/Make | No-code | Self-hosted n8n |
| Need custom logic in >50% of workflows | No-code | Low-code or code |
| >100K executions/day | Any hosted | Self-hosted or custom |
| Complex branching breaking visual tools | Low-code | Custom code |
| Multiple teams building automations | Single tool | Platform + governance |
| AI judgment needed in workflows | Traditional | AI agent integration |
Every automation must be registered:
automation_registry_entry:
id: "WF-[DEPT]-[NUMBER]"
name: "[Descriptive name]"
description: "[What it does in one sentence]"
owner: "[Person]"
team: "[Department]"
platform: "[n8n/Zapier/Make/custom]"
status: "[active/paused/deprecated/testing]"
created: "[date]"
last_modified: "[date]"
last_reviewed: "[date]"
review_frequency: "[monthly/quarterly]"
business_impact:
time_saved_monthly_hours: [X]
cost_saved_monthly: "$[X]"
error_reduction: "[X%]"
technical:
trigger: "[type]"
systems_connected: ["system1", "system2"]
avg_daily_executions: [X]
success_rate: "[X%]"
dependencies:
upstream: ["WF-XXX"]
downstream: ["WF-YYY"]
documentation:
blueprint: "[link]"
runbook: "[link]"
test_plan: "[link]"
Pattern: [DEPT]-[ACTION]-[OBJECT]-[QUALIFIER]
Examples:
SALES-sync-leads-from-typeform
FINANCE-generate-invoice-monthly
HR-onboard-employee-new-hire
MARKETING-post-content-social-scheduled
OPS-backup-database-nightly
| Change Type | Approval | Testing | Rollback Plan |
|---|---|---|---|
| Config change (threshold, timing) | Owner | Quick smoke test | Revert config |
| Logic change (new branch, new step) | Owner + reviewer | Full test suite | Previous version |
| Integration change (new API, new system) | Owner + tech lead | Integration + E2E | Disconnect + manual |
| New workflow | Owner + stakeholder | Full test + pilot | Disable workflow |
| Deprecation | Owner + affected teams | Verify replacements | Re-enable |
| Scenario | AI Type | Example |
|---|---|---|
| Classify unstructured text | LLM | Categorize support tickets |
| Extract data from documents | LLM + OCR | Parse invoices, contracts |
| Generate content from templates | LLM | Personalized emails, reports |
| Make judgment calls | LLM + rules | Lead scoring, risk assessment |
| Summarize information | LLM | Meeting notes, research briefs |
| Route based on intent | LLM | Customer request → right team |
ai_agent_step:
type: "ai_judgment"
model: "[model name]"
input:
context: "[relevant data from previous steps]"
task: "[specific instruction — be precise]"
output_format: "[JSON schema or structured format]"
constraints: ["must not", "must always", "if unsure"]
validation:
confidence_threshold: 0.85
required_fields: ["field1", "field2"]
value_ranges:
score: [0, 100]
category: ["A", "B", "C"]
on_low_confidence:
action: "route_to_human"
queue: "[review queue name]"
on_failure:
action: "fallback_to_rules"
rules_engine: "[rule set name]"
monitoring:
log_all_decisions: true
sample_rate_for_review: 0.10
alert_on_confidence_drop: true
| Level | Name | Description | Indicators |
|---|---|---|---|
| 1 | Ad Hoc | Manual processes, maybe a few scripts | No registry, tribal knowledge |
| 2 | Reactive | Automate pain points as they arise | Some workflows, no standards |
| 3 | Systematic | Planned automation program | Registry, testing, monitoring |
| 4 | Optimized | Continuous improvement, governance | ROI tracking, quarterly reviews |
| 5 | Intelligent | AI-augmented, self-healing | Adaptive workflows, predictive |
automation_maturity:
dimensions:
strategy: [1-5] # Planned roadmap vs ad hoc
architecture: [1-5] # Patterns, standards, reuse
reliability: [1-5] # Error handling, monitoring, uptime
governance: [1-5] # Registry, change management, reviews
testing: [1-5] # Test coverage, validation, chaos
documentation: [1-5] # Blueprints, runbooks, training
optimization: [1-5] # Performance, cost, continuous improvement
ai_integration: [1-5] # AI-powered decisions, self-healing
total: [sum ÷ 8]
grade: "[A/B/C/D/F]"
# A: 4.5+ | B: 3.5-4.4 | C: 2.5-3.4 | D: 1.5-2.4 | F: <1.5
top_gap: "[lowest scoring dimension]"
next_action: "[specific improvement for top gap]"
| Dimension | Weight | 0-2 (Poor) | 3-5 (Basic) | 6-8 (Good) | 9-10 (Excellent) |
|---|---|---|---|---|---|
| Design | 15% | No blueprint, ad hoc | Basic flow documented | Full blueprint with error handling | Blueprint + edge cases + optimization |
| Reliability | 20% | No error handling | Basic retries | DLQ + circuit breaker + fallback | Self-healing + auto-scaling |
| Testing | 15% | No tests | Happy path only | Full test pyramid | Chaos testing + load testing |
| Monitoring | 15% | No visibility | Basic success/fail logs | Dashboard + alerts | Predictive monitoring |
| Documentation | 10% | None | README exists | Blueprint + runbook | Full docs + training materials |
| Security | 10% | Hardcoded credentials | Encrypted secrets | Least privilege + rotation | Zero-trust + audit trail |
| Performance | 10% | Works but slow | Acceptable speed | Optimized + cached | Auto-scaling + sub-second |
| Governance | 5% | No registry | Listed somewhere | Full registry + reviews | Change management + compliance |
Score: (weighted sum) → Grade: A (90+) B (80-89) C (70-79) D (60-69) F (<60)
| # | Mistake | Fix |
|---|---|---|
| 1 | Automating a broken process | Fix the process FIRST, then automate |
| 2 | No error handling | Every step needs a failure path |
| 3 | Silent failures | If it fails and nobody knows, it's worse than manual |
| 4 | Not testing edge cases | Test empty, duplicate, malformed, concurrent |
| 5 | Hardcoded values | Use config/environment variables for everything |
| 6 | No monitoring | You can't fix what you can't see |
| 7 | Building monolith workflows | One workflow, one job. Chain them together |
| 8 | Ignoring rate limits | Design for API limits from day one |
| 9 | No documentation | Future-you will hate present-you |
| 10 | Over-automating | Not everything should be automated. Human judgment exists for a reason |
Use these to invoke specific phases:
audit my processes for automation opportunities → Phase 1prioritize automations by ROI → Phase 2recommend automation platform for [process] → Phase 3design workflow blueprint for [process] → Phase 4plan integration between [system A] and [system B] → Phase 5design error handling for [workflow] → Phase 6create test plan for [automation] → Phase 7set up monitoring for [workflow] → Phase 8optimize [workflow] for scale → Phase 9review automation governance → Phase 10add AI to [workflow] → Phase 11assess automation maturity → Phase 12