XRefKit.

THE OPERATING BASE FOR AI WORK

From AI work.
To accepted
outcomes.

Connect Goals, Skills, Knowledge, and Evidence in a traceable workflow. Keep the next responsibility clear—and acceptance in human hands.

Start with XRefKit Explore the model
REPOSITORY-NATIVESTABLE REFERENCESHUMAN ACCEPTANCE
FOLLOW THE EVIDENCEILLUSTRATIVE EXAMPLE
OUTPUT / C# REVIEW

A finding you can inspect.

The finding, its rationale, and unresolved items travel together.

01Trace the referencesSkill + selected Knowledge

Which procedure and criteria informed this finding?

C# review Skill ↗Selected Knowledge ↗

Used XIDs connect the record to stable references.

02Inspect the checksRecord verification + output review

Check the execution record. Review output validity separately when required. Keep unresolved items visible.

Explore reporting and evidence ↗
03Keep the decision humanAcceptance against the Goal

A person inspects the output and Evidence against acceptance conditions, then accepts, returns, stops, or routes the work onward.

Explore human acceptance ↗
One output. Its references, checks, and decision context.
A schematic example, not a live execution record.
Built around a simple ideaOrganizations scale by replicating judgment.

Current release

Latest published release: v0.6.3 (). Development notes on the linked pages describe work in the repository; they do not imply availability in that release.

Explore what is shipping and what comes next.

1. The common foundation and its ongoing development2. Individual Skills and the work they perform

ONE FOUNDATION. DIFFERENT KINDS OF WORK.

Find your way in.

01 / EXPLORE

Common foundation

Shared reference delivery, execution records, checks, and handoff support every kind of work. Current development connects accepted output quality, plan views, and an Azure DevOps integration trial.

Read the foundation and current development →

02 / EXPLORE

Common protocols

Workflow progression, reporting, output quality, uncertainty and authority, Knowledge and execution context, handoff, and shared-asset updates. Each contract has its own applicability conditions.

Explore the shared contracts →How reporting works →
03 / EXPLORE

Individual Skills

Skills define the method and outputs for a particular job. Examples include C# review, constraint derivation, and business learning through interviews. They use the common foundation while keeping their own scope and criteria.

Explore individual Skill examples →See the C# responsibility split →

Operating model

People define and accept the result. AI performs bounded work. Shared controls preserve the record and support review.

Who decides what — an overview of the operating model
  1. Human · define

    Goal and acceptance criteria

    Set the desired business result and the conditions for accepting it.

  2. AI · execute

    Route and run a bounded Skill

    Use the selected Knowledge and preserve outputs, decisions, and unresolved items as Evidence.

  3. Controls · verify and review

    Check records and output quality

    Verify the workflow record and review actual output quality separately when required. Prepare handoff and close after the required checks.

  4. Human · accept

    Decide adoption and the next request

    Inspect the output, evidence, and handoff. Accept the result or request repair.

Escalation → Human judgment: when work crosses the current authority boundary, return the decision to people.

Stable XIDs connect the selected Skills and Knowledge to the execution record. This overview groups responsibilities; the detailed model defines each element.

Explore the operating model and its 12 elements →

Get started

Python 3.11 or later is required. Install XRefKit, then initialize it in your target workspace.

1. Install
python -m pip install xrefkit
2. Initialize your workspace
xrefkit init
xrefkit --help
3. Connect an AI client if needed

MCP is optional. Follow the repository's setup guide for your client and review the generated configuration before applying it.

Read the setup and MCP instructions