Case study

Project Spine

An open-source context compiler: brief + repo + design tokens in, a deterministic, drift-checked operating layer for humans and coding agents out.

toolingOpen sourceCLIAI workflowsDesign tokensMCP

Live site Full case study on Digitaltableteur

Estimated reading time

Key metrics

Releases

April to June 2026, alpha through beta

Files per compile

agent instructions, guardrails, plans, backlog

21+

First run

enforced by an executable readiness check

<30s

Outcome

Three signals of impact.

14 releases in three months (v0.1.0-alpha to v0.9.2-beta.2), published to npm with provenance and an executable release-readiness gate
One compile writes 21+ files: instructions for four agent tools, QA guardrails, scaffold plan and a sprint-1 backlog, with every rule carrying a source pointer
Drift detection ships as a CI gate with a reusable GitHub Action, so instruction files stay aligned as the repo evolves

Role and scope

Creator and maintainer: product, architecture, CLI and MCP implementation, positioning

Constraints

Deterministic before enriched: the optional LLM call is additive prose only and never load-bearing, because LLM contamination breaks drift detection by definition
Useful without AI, repo-native first, and a first run enforced under 30 seconds by an executable readiness check
Pure OSS surface: hosted-tier code exists but is unrouted and excluded from the published package by a pack check

Top decisions

Every compiled rule carries provenance back to the brief, repo profile or template that produced it
Drift is a CI gate with exit codes, not a report
Design tokens accepted as DTCG or Tokens Studio JSON, with Figma Variables pull where the plan allows

Confidential details redacted where required.

Next step

If you want this outcome, we can scope the engagement quickly.

Contact Work with me