返回 Skill 列表
extension
分类: AI Agent 能力无需 API Key

SDA Controller

模拟-部署-增强循环取代传统的PDCA

person作者: jakexiaohubgithub

SDA Controller Skill

Replaces traditional PDCA (Plan-Do-Check-Act) with a faster, AI-native iteration loop.

SDA vs PDCA

| PDCA | SDA | Improvement | |------|-----|-------------| | Plan | Simulate | Test before building | | Do | Deploy | Canary first | | Check | Augment | Real-time learning | | Act | (Continuous) | No manual cycle |

SDA Loop

┌─────────────────────────────────────────────┐
│                                             │
│  SIMULATE ──► DEPLOY ──► AUGMENT            │
│      ▲                       │              │
│      └───────────────────────┘              │
│                                             │
└─────────────────────────────────────────────┘

Phase Definitions

1. SIMULATE

Run solution in high-fidelity sandbox before any real deployment.

simulate:
  environment: sandbox
  actions:
    - Run test suite: `npm run test`
    - Build check: `npm run build`
    - Load test: `npm run load-test:smoke`
  success_criteria:
    tests_pass: true
    build_success: true
    performance_threshold: "p95 < 200ms"

2. DEPLOY

Small-scale canary release with rollback capability.

deploy:
  strategy: canary
  initial_traffic: 5%
  ramp_schedule:
    - 5%: 10min
    - 25%: 30min
    - 50%: 1hr
    - 100%: 2hr
  rollback_trigger:
    error_rate: "> 1%"
    latency_p99: "> 500ms"

3. AUGMENT

Agents learn from telemetry, adjusting in real-time.

augment:
  telemetry_sources:
    - Error logs
    - Performance metrics
    - User feedback signals
  feedback_actions:
    - Auto-fix common patterns
    - Generate hotfix PRs
    - Update configuration
  learning_loop:
    - Store successful patterns
    - Avoid failed approaches

Workflow Commands

/iterate simulate    # Run simulation only
/iterate deploy      # Deploy with canary
/iterate full        # Complete SDA cycle
/iterate rollback    # Emergency rollback

Integration Points

| Component | Purpose | |-----------|---------| | eval-harness | Powers SIMULATE phase | | verification-loop | Continuous checks | | GitHub Actions | DEPLOY automation | | Sentry | AUGMENT telemetry |

Success Metrics

| Metric | Target | |--------|--------| | Simulation pass rate | 100% | | Canary success rate | > 99% | | Mean time to augment | < 5min | | Rollback frequency | < 1% |