---
title: Agentic Pipeline Runner
slug: agentic-pipeline-runner
type: system
created: 2026-06-17
updated: 2026-06-17
status: active
tags:
  - runner
  - pipeline
  - agents
  - dot
sources:
  - strongdm-attractor-nlspec
  - 2389-dark-factory-dot-file-2026-03-09
  - jleechanorg-dark-factory-repo
related:
  - dot-pipeline
  - attractor-pattern
  - agent-orchestrator
  - cxdb
  - healer
confidence: high
contradictions:
  - The runner can be considered disposable relative to the graph, but runner correctness is still load-bearing for safety and reproducibility.
bead_refs:
  - dark-factory-rr4
---

# Agentic Pipeline Runner

An agentic pipeline runner executes a [[dot-pipeline]] by parsing the graph, validating structure, invoking node handlers, routing outcomes, recording events, and coordinating agents or tools.

## Expected Responsibilities

- Parse graph source into nodes, edges, attributes, and routing conditions.
- Validate unsupported shapes, missing prompts, invalid gates, and unsafe graph constructs.
- Initialize execution context: workdir, backend, feature, run ID, CXDB, evidence bundle, and environment.
- Execute deterministic nodes, LLM/coding-agent nodes, gates, human gates, joins, retries, and exits.
- Finalize run status and emit evidence for debugging and audit.

## Concrete Source

The [[jleechanorg-dark-factory-repo]] is the concrete implementation source in this wiki. It includes Python runner code, prompt catalogs, DOT pipelines, CXDB logging, Healer diagnosis, benchmark projects, and backend adapters for multiple agent systems.

## Boundary

[[agent-orchestrator]] is not the same thing. It manages workers, sessions, worktrees, and PR feedback loops. A pipeline runner manages a process graph and may call an orchestrator as one backend.