3.1 Pipeline Stages

Pipeline Stages

The Pipeline Engine (~/system/kernel/pipeline-engine.js) manages a multi-company chain where each completed stage triggers the next.

Stage Flow

BUILD (CodeCraft / Vizu / Datavera)
  │
  ▼  Task completed → Pipeline Engine creates REVIEW task
REVIEW (Proveo)
  │
  ├── Pass    → Pipeline Engine creates SECURITY task
  └── Reject  → New BUILD task with findings (max 2 cycles)
        │
        ▼
SECURITY (Securion)
  │
  ├── Pass    → Pipeline Engine creates OPS task (if deploy)
  └── Reject  → New BUILD task with security findings
        │
        ▼
OPS (FlowForge) — optional, triggered for deploy tasks
  │
  ▼
DOCS (Lexicon) — optional, triggered for docs tasks

How It Works Technically

1. Pi-Orchestrator completes a BUILD task
2. Calls pipeline-engine.js advance <task-id>
3. Pipeline Engine checks the task's pipeline stage
4. Creates a new MC task for the next stage with:
   • Company context injected
   • Previous stage output as input
   • Read-only flag for REVIEW/SECURITY (they can't modify code)
5. Pi-Orchestrator picks up the new task in the next cycle

Pipeline Metadata

Each task in a pipeline has metadata tracking:

• pipeline_id — groups related tasks
• pipeline_stage — current stage (BUILD/REVIEW/SECURITY/OPS/DOCS)
• pipeline_parent — ID of the task that triggered this one
• review_cycle — current review iteration (max 2)

Safety: Max Review Cycles

To prevent infinite BUILD↔REVIEW loops, the pipeline caps at 2 review cycles. After that, the task is queued for human review.