| name | orchestrator |
| description | Use when you need to execute a multi-step pipeline of finops roles with defined order, dependencies, and validation gates between stages |
Pipeline Orchestrator Skill
Overview
Executes a pipeline of FinOps roles in defined order, passing context between stages, and enforcing hard gates at each stage boundary.
When to Use
- Complex task requiring multiple roles in sequence
- Task with parallel branches that must converge
- Task where each stage output feeds into the next
Execution Rules
Stage Order
- Execute stages in defined order
- If stage has
depends, wait for dependencies to complete
- If stage has
parallel, execute all in parallel, wait for all to complete
Data Passing
- Reference previous stage output:
stage://stage-name/variable-name
- All stage outputs stored in pipeline context
- Missing reference → BLOCK at gate
Gate Rules
| Condition | Action |
|---|
| Previous stage output missing | BLOCK, require retry |
| Required field empty | BLOCK, require 补充 |
| Stage execution error | Allow 2 retries, then BLOCK |
FinOps Pipeline
stages:
- route-intent: 路由到正确角色
- strategy-formulation: 策略制定（串行）
- execution: 策略/内容/活动/产品并行执行
- compliance-check: 合规审查（必须在上线前）
- data-review: 数据复盘
FinOps-Specific Rules
- 合规线永远在发布前最后一步: 不提前、不跳过
- 运营产出必须过合规闸: 不允许先发后补
- 数据线是监控和反馈: 不直接触达用户
Hard Gates
- Pre-Stage Gate: Verify all inputs available before stage execution
- Post-Stage Gate: Verify stage output complete before proceeding
- Compliance Gate: Verify compliance-ops has approved before publishing
Anti-Patterns
- Skipping a stage because "it's obvious what to do"
- Proceeding when a previous stage output is missing
- Skipping compliance gate before publishing
- Hardcoding stage order instead of reading pipeline.yaml
