Product lifecycle for AI.

Agents do better work with context they can trust. HELIX is a document discipline for teams building software with agents. It turns project intent and evidence into shared memory, then keeps that memory current as the work changes.

Start with HELIX Inspect the proof

A document spine double helix connecting human intent, AI execution, and governed artifacts.

Get started

Install HELIX in your runtime

HELIX works with Claude Code, Codex CLI, GitHub Copilot, and Databricks Code Genie. Pick the guide for your environment and you’re up in minutes.

Claude Code Install as a Claude Code skill User-global or per-repo. Supports every HELIX route including build and run. Codex CLI + Copilot Discover via AGENTS.md or editor instructions Full flow on the Codex CLI; read-mostly HELIX work on Copilot surfaces. Databricks Code Genie Skill bundle in workspace storage Shared across the workspace, next to your data and notebook work.

Read the full install index

How it works

Getting started with HELIX is simple: install it into your agent, then work through prompts as you normally would. The HELIX skill helps the agent read and update the document graph while you stay in control. That graph gives the agent the context it needs to create, execute, and refine quality plans.

1 · Write the brief Say what the agent should know. Capture the goal, requirements, constraints, and decisions before the agent starts changing files. 2 · Check alignment Find what no longer matches. Use HELIX to catch stale assumptions, missing context, and contradictions across the document graph. 3 · Create the work plan Turn the documents into bounded work. Define what to change, what not to change, how to check success, and what evidence to collect. 4 · Run it in the factory Send the plan to the place work gets done. Use DDx, Claude, Codex, Databricks, or a manual workflow. Capture the result and feed it back into the documents.

See it in action

Artifact spine

HELIX includes many artifact types, but to get started you only need a few. Begin with the spine, then add supporting artifacts as the work demands them.

01Product VisionIntent 02PRDRequirements 03PrinciplesJudgment 04Feature SpecsScope 05ArchitectureStructure 06Test PlansProof 07Implementation PlansExecution

Worked example: HELIX governs itself

The graph below shows HELIX's own governing artifacts on the same spine. Hover each document to see what changed and the kind of prompt used in the public screencasts.

Human judgment Manual edits + prompts

AI execution Alignment reviews + updates

01 Intent Product Vision Why HELIX exists. The frame prompt created the first intent document before requirements or code. Create the HELIX product vision. 02 Requirements PRD What HELIX provides and refuses. The same frame prompt turned intent into concrete product requirements. Create the HELIX PRD. 04 Scope Feature Specs Capabilities derived from requirements. The frame prompt decomposed the PRD into acceptance-backed feature work. Create feature specs from the HELIX PRD. 03 Constraint Principles Values for design and review. Review prompts use the governing documents to catch omissions and drift. Review HELIX against its governing principles. 05 Structure Architecture Boundaries and system shape. The design prompt translated scoped requirements into a technical shape. Create the HELIX architecture. 06 Proof Test Plans Executable expectations. The test prompt asked for red tests before implementation touched the code. Create test plans for the HELIX feature specs. 07 Execution handoff Implementation Plans Scoped runtime work. The implementation prompt sent bounded work to the runtime with a test gate. Create a HELIX implementation plan.

Humans provide judgement and quality, steering with edits and prompts. Agents review, propose updates to keep the artifact stack coherent, create implementation plans and execute them.

Runtime paths

Use HELIX where your team already works

HELIX is Markdown and methodology. The runtime supplies file editing, review, execution, and evidence capture.

Manual Small teams adopting the method first Use Markdown, reviews, and explicit prompts before adding queue automation. Claude Code Interactive artifact review and editing Ask an agent to reconcile documents, propose patches, and explain open questions. Codex Codebase-aware planning and implementation Use the artifact spine to guide bounded code changes and implementation plans. DDx Reference runtime for queued execution Map HELIX plans to beads when queue control and execution evidence matter. Databricks Data and governance workflows Apply HELIX artifacts to analytical systems, platform work, and managed evidence.

Proof

Inspect the foundations

The method is public: the catalog, HELIX’s own governing artifacts, and the research foundation are all inspectable.

Worked artifacts HELIX governs itself in public Read the actual vision, PRD, principles, contracts, designs, and alignment reviews behind the project. Artifact catalog Reusable prompts, templates, and quality guidance Browse the document types and learn which ones are core versus supporting. Research foundations Why document-driven AI work needs governance Trace the methodology back to the research and operating assumptions that shaped it.