Pipeline Architecture

The pipeline manages issues through a structured lifecycle with feedback loops for quality assurance.

Overview

SCOPE ──▶ BUILD ──▶ REVIEW ──▶ TEST ──▶ DONE
                       ▲         │
                       └─────────┘
                    (feedback loops)

For detailed stage documentation, see Pipeline Stages.

Pipeline Data Model

@dataclass
class Pipeline:
    id: str
    project_id: str
    linear_issue_id: str
    stage: PipelineStage        # scope, build, review, test, done, failed
    branch_name: str | None
    worktree_path: str | None
    pr_url: str | None
    scope_result: str | None    # JSON ScopeResult
    review_result: str | None
    test_report: str | None
    qa_steps: str | None        # JSON QA steps

Stage Transitions

Current	Signal	Next
SCOPE	scope-complete	BUILD
BUILD	build-complete	REVIEW
REVIEW	review-approved	TEST
REVIEW	review-changes-requested	BUILD
TEST	test-passed	DONE
TEST	test-failed	REVIEW

PipelineRunner

The PipelineRunner service:

1. Watches for signal files in worktrees
2. Detects stage completion signals
3. Spawns the next agent
4. Updates Linear state
5. Handles retries on failure

# Polling interval
POLL_INTERVAL_SECONDS = 5

# Max retries before FAILED
MAX_RETRIES = 3

Linear State Mapping

Pipeline Stage	Linear State
SCOPE	In Progress
BUILD	In Progress
REVIEW	Code Review
TEST	QA
DONE	Done

Commands

goblin pipeline start ENG-123   # Start pipeline
goblin pipeline status ENG-123  # Check status
goblin pipeline retry ENG-123   # Retry failed
goblin pipeline cancel ENG-123  # Cancel

Next Steps

• Pipeline Stages - Detailed stage documentation
• Signal Files - Signal file reference
• Storage - Database schema