AX Engineering Console

Docs

The AX Engineering Harness is a reusable engineering layer for reliable AI-assisted software development: context, roles, procedures, controlled capability, verification and evaluation, kept in the repository rather than in a prompt.

It is an experimental learning project. The contracts and their validation are real; execution is not. Nothing in it runs a command, starts an agent, or enforces a permission.

Quick start

The Phase 13 CLI is available as ax after using the repository installer. It builds and links the CLI from source; the packages are still private and are not published to npm.

  • ax initPropose an adoption for the current directory. Writes nothing.
  • ax init --writeCreate the files it listed. An existing file is never replaced.
  • ax doctorReport what is declared, what exists on disk, and what could run.
  • ax migrateRename a Claude workspace generated before the commands carried the ax- prefix. Writes nothing without --write.
  • ax validateResolve a project declaration and check its context references.
  • ax qualityPlan the gates for a project. Runs none of them.
  • ax policyPrint the capability matrix from the definitions.
  • npm run checkBuild, typecheck, lint, test every package, and build the console.

What’s new: CLI distribution

The CLI now lives in its own package beside the shared core and web console. Installers build and link it locally, and the Claude Code starter pack is supplied through an adapter rather than imported by the vendor-neutral core. This makes the harness easier to try in another repository while keeping adoption preview-first: ax init writes nothing unless you pass --write.

Distribution is experimental: packages remain private; no npm release is claimed.

Topics

Each topic has a page in this console and an authoritative document in the repository. The document is the source of record; this console renders the definitions those documents describe.

  • Architecturedocs/visualizations.md

    How a request moves from a person to a reviewed change.

  • Workflow and handoffdocs/workflow.md

    Lifecycle states, permitted transitions, bounded retries, and what a handoff must record.

  • Project configurationdocs/project-configuration.md

    The declaration contract: fields, limits, and what validation does not authorise.

  • Project inventorydocs/project-inventory.md

    Which areas a project has, from known file names only, and the bounds that keeps discovery safe.

  • Profilesdocs/profiles.md

    Stack assumptions, build detection, and what each profile refuses to assume.

  • Agents and skillsdocs/agents-and-skills.md

    Roles and procedures, and why an agent is not a skill.

  • MCP and toolsdocs/adapters.md

    The interface layer, and the vendor entrypoints that reach the shared instructions.

  • Permissionsdocs/policies.md

    The capability vocabulary, what requires approval, and what is denied to everyone.

  • Quality gatesdocs/quality-and-evals.md

    The pipeline, and why unavailable and unrun are not passes.

  • Evalsdocs/quality-and-evals.md

    Assessing agent behaviour rather than software.

  • AdoptionREADME.md

    Applying the harness to an existing repository.

What to expect

  • Definitions, schemas and validation are implemented and tested.
  • Roles, procedures, gates and evals are defined; no runtime carries them out.
  • MCP connections, gate execution and real runs are planned, not built.
  • Every record and score displayed here is an example, labelled where it appears.