ian-rainford

AI-assisted development

Helped introduce AI-assisted development with Claude (Claude Code) into the squad's day-to-day workflow, speeding up delivery across the Kotlin/Spring Boot and TypeScript/React codebase.

How a change reaches main

  1. step 01 / 06

    Load the project context

    Project context files such as CLAUDE.md capture the architecture, conventions and build and test commands, so generated code follows team standards from the start.

  2. step 02 / 06human checkpoint

    Plan first

    Work starts with a plan, not code. A person stays in the loop and agrees the approach before anything is implemented.

  3. step 03 / 06

    Small, reviewable steps

    Implementation happens in small steps that are easy to review, instead of one large change.

  4. step 04 / 06

    Same pipeline, same gates

    AI-generated changes run through the same CI pipeline and test gates as hand-written code. AI also helps generate the tests.

  5. step 05 / 06human checkpoint

    Peer review

    A teammate reviews the pull request, reading AI output as critically as any other code.

  6. step 06 / 06human checkpoint

    Approved, then merged

    Only reviewed changes are merged. Output from custom skills goes through the same pull request process.

square steps are human checkpoints

AI-assisted development

  1. 01

    Work in an agentic, human-in-the-loop style: plan first, then implement in small, reviewable steps. Every AI-generated change goes through the same peer review, CI pipeline and test gates as hand-written code.

  2. 02

    Practise context and prompt engineering: keep project context files (such as CLAUDE.md) that capture architecture, conventions and build/test commands, so generated code follows team standards.

  3. 03

    Use AI for test generation, legacy refactoring, debugging and API/documentation drafting, freeing time for design and technical decisions.

  4. 04

    Coach teammates on responsible use: writing effective prompts, reviewing AI output critically, and protecting sensitive code and data.

Custom Claude skills and workflow automation

Designed and built custom Claude skills (reusable, version-controlled instruction packages) that encode team workflows, so routine tasks run the same way for everyone.

  • /api-security-review

    API security review skill

    Checks endpoints against the OWASP API Security Top 10, covering authentication and authorisation, input validation, error handling and data exposure, and produces findings for human review.

  • /feature-toggle-lifecycle

    Feature toggle lifecycle skill

    Audits flags, identifies stale toggles after rollout, and drafts the clean-up changes that remove the flag and its dead code paths.

  • /security-patching

    Security patching skill

    Triages dependency vulnerability alerts (CVEs), proposes upgrades, runs the build and tests, and prepares pull requests for review.

// noteHumans stay in the loop: skill output is reviewed and approved through the normal pull request process.