Claude Update Wizard: Guided, Selective Skill Updates

Where this fits

You've been running the claude-sdlc-harness wizard for a few weeks. You've added custom step templates, tweaked prompts, and wired it into your team's workflow. A new version lands on GitHub. Your agent's first instinct might be to pull the repo and overwrite everything — which wipes your customizations. The claude-update-wizard skill turns that risky update into a guided, diff-first process: it reads your installed version, generates a changelog, compares each file against upstream, and lets you adopt changes selectively before committing anything.

Why agents benefit

• Reads the installed version from SDLC.md metadata and maps it to the correct update path instead of guessing.
• Produces a real changelog before touching files, so the agent and user see exactly what changed between versions.
• Diffs file-by-file, identifying which upstream changes conflict with local customizations.
• Supports check-only and force-all modes, giving the agent a way to inspect first or commit wholesale only when the user explicitly asks.
• Follows an ordered execution checklist, preventing agents from combining or skipping verification steps.

Practical scenarios

Preserving custom step templates

When the wizard adds a new step registry, the update agent needs to know your previously completed steps are encoded in SDLC.md. It reads the Completed Steps comment, merges new entries, and shows you exactly which step definitions changed — so you can keep your custom ones without breaking step dependencies.

Auditing before a forced update

A change in the harness's prompt format might be breaking. Running with the check-only argument makes the agent read the changelog and diffs and report back "safe to update" or "three files need attention" — no writes, no surprises, no rollback choreography.

Team-wide consistent upgrades

Multiple agents in a repo share the same SDLC.md. The wizard's idempotency means the agent can read the version comment and skip re-registering unchanged steps. Your team gets uniform updates without one agent overwriting another's configuration mid-sprint.

Add it to your agent workflow

Install it from the CLI:

npx skills add BaseInfinity/claude-sdlc-harness --skill claude-update-wizard

Then invoke it inside a session where the SDLC wizard is already present:

/claude-update-wizard check-only

The agent reads CLAUDE_CODE_SDLC_WIZARD.md first, parses the version from SDLC.md, fetches the remote changelog, and walks you through each diff — you approve or reject adoption per file. For more context on how this skill fits the broader harness, check the entry on OpenAgentSkill.

Compare before adopting

Before letting the wizard update, compare: the quality of the changelog (is it descriptive enough to make an informed call?), maintenance freshness (are updates regular or sporadic?), alternatives like a manual git pull plus patch, and workflow fit — does your team standardize on the SDLC harness or drift toward other tooling?

Why it is worth tracking

claude-update-wizard sits inside a small but active ecosystem: the SDLC harness has 44 stars, and the wizard's design — read, diff, decide — signals a mature approach to safe agent-driven maintenance. Track it if you rely on wizard-style agents that undergo frequent upgrades; it's a good benchmark for how agent skills can handle versioned, stateful tooling that must preserve user customizations rather than blindly overwrite them.