dot-files

A collection of mammoth DOT pipelines for use with tracker — the agentic pipeline engine that executes multi-step AI workflows defined as Graphviz DOT graphs.

Quick Start

# Install tracker
go install github.com/2389-research/tracker/cmd/tracker@latest

# Configure API keys
tracker setup

# Run a pipeline
tracker speedrun.dot

You will need API keys for the LLM providers used in each pipeline (Anthropic, OpenAI, Gemini). See the tracker docs for details.

Pipelines

Development Pipelines

Pipeline	Description
speedrun.dot	Ultra-minimal build pipeline — fastest path from spec to shipped code. No brainstorm, no design doc. Read spec, plan, implement, test, ship. Fully headless.
bug-hunter.dot	Autonomous bug fix pipeline — reads a bug report, reproduces, diagnoses, fixes via TDD, and ships a PR. Fully headless.
refactor-express.dot	Autonomous incremental refactoring — analyzes code, plans steps where tests stay green at every step, executes with rollback on failure.
doc-writer.dot	Autonomous documentation generator — explores a codebase and produces README, API reference, architecture guide, and tutorials with verified code examples.

Interactive Pipelines

Pipeline	Description
20q.dot	20 Questions game — the AI asks yes/no questions to guess what you're thinking of within 20 questions.
story-engine.dot	Choose-your-own-adventure game — AI writes branching narrative scenes, you make story choices that shape the plot across multiple chapters.
model-debate.dot	Multi-model debate arena — Claude, GPT, and Gemini argue positions on a topic you choose across three rounds, then you judge the winner.

How It Works

Each .dot file defines a directed graph where nodes represent pipeline steps and edges define control flow. Node shapes map to handler types in tracker:

• box — LLM call with tool access
• hexagon — Human input gate (yes/no or freeform)
• parallelogram — Shell command execution
• component / tripleoctagon — Parallel fan-out / fan-in
• diamond — Conditional branching

Pipelines configure LLM providers and models via model_stylesheet in the graph attributes. Most pipelines use a mix of Claude, GPT, and Gemini models for different roles.

Related Projects

• tracker — The runtime engine that executes these pipelines
• mammoth — The DOT pipeline format and ecosystem
• dotpowers — Superpowers-based pipelines (kitchen-sink, test-kitchen, scenario-testing)

---

Built by 2389.ai