Back to selected work

Engineering enablement · AI/LLM · Human review

Reusable AI workflows for software delivery.

I built reusable Claude Code skills and project plugins that connected engineering work to Jira. Tests, pull-request builds and code review remained required.

Role
AI workflow design, plugin development, adoption and review standards
Technology
Claude Code · Atlassian MCP · Jira · Project plugins · CI

01 / Background

Context

Engineers repeatedly spent time supplying project architecture, domain knowledge, coding conventions and delivery workflows before an AI assistant could produce useful work. Creating tickets also meant manually translating implementation details into a consistent Jira structure.

02 / Constraints

Constraint

The tooling needed explicit project knowledge, installable team workflows and the same engineering checks used for manually produced changes. Speed could not come at the expense of accuracy or review.

03 / Change

Before / after

Before

Project context and ticket structure had to be rebuilt repeatedly, slowing first passes and increasing the chance of context-related implementation mistakes.

After

Reusable skills and project plugins supplied project-specific context, created structured tickets directly in Jira and produced work for engineers to review.

04 / Architecture

System design

  1. 01

    Encoded project context

    Architecture, domain knowledge, conventions and delivery workflows

  2. 02

    Reusable AI tooling

    Claude Code ticket skill, repository plugins and Atlassian MCP

  3. 03

    Engineering verification

    Unit tests, pull-request builds and human code review

05 / Decisions

Technical decisions

  1. 01

    Include project context

    Package project architecture, domain language, coding conventions and development workflows so useful constraints travelled with the tool.

  2. 02

    Connect tools to the delivery workflow

    Generate structured development tickets and create them directly in Jira through Atlassian MCP instead of stopping at disconnected draft text.

  3. 03

    Keep the existing review process

    Validate AI-assisted changes through unit tests, pull-request builds and human code review before treating generated work as complete.

06 / Leadership

My contribution

I developed Claude Code plugins for two production codebases, documented how other developers could install them and kept engineers responsible for reviewing the resulting work.

07 / Results

Outcome

The reusable tooling improved AI-assisted development speed, reduced context-related implementation mistakes and returned more engineering time to architecture, edge cases and review across two production codebases.

Next case study

Legacy B2B commerce modernisation

Read next