Wholesale AI Champions Claude Code playbook
← Champions

Kyle Trauberman

Windows · .NET · Azure DevOps · ADESA Untitled team

Draft — Claude assembled this from Kyle’s actual ~/.claude configs. Sanity-check, edit voice, and rewrite anything that doesn’t ring true before merging.

Settings

  • I use ccstatusline for my status line in Claude Code.
  • The permission philosophy in settings.local.json is “verbose allow-list for read-only, hard deny on anything destructive”.

Plugins

Carvana (carvana-claude-code-marketplace). essentials — the Pathfinders multi-skill workflow (brainstorm → analyze → plan → breakdown → tdd → execute → check → review → reflect). Whenever I run an essentials:* skill against a PBI my project CLAUDE.md tells Claude to also tag the PBI RPIV in ADO so Pathfinders can track adoption.

Wholesale (wholesale-claude-code-marketplace). wholesale-backlog-builder (decompose a PM spec into ADO Features + PBIs with vertical slicing and Figma context, then upload), and ws-atlas (the CVNA-Wholesale ecosystem graph — 376 repos, 58 apps, 14 teams — for ownership lookups, alert triage, and blast-radius analysis).

MCPs

Connected globally in ~/.claude.json:

  • Context7 — current library docs, so Claude doesn’t lean on stale training data when I’m in unfamiliar packages. Stdio: npx -y @upstash/context7-mcp.
  • Azure DevOps — the official @azure-devops/mcp server bound to org adesacentral, scoped via -d to the core, work, work-items, search, and pipelines domains. My global CLAUDE.md pins the defaults: project adesa, area path Adesa\Untitled, iteration ADESA, parent feature 155843 for any new work item.
    • ~/.claude/claude.md contains defaults for the azure devops server, including the tenant and project name, as well as a default area, iteration, and parent feature for new PBIs.

Install commands for the cross-team MCPs live on the MCPs page.

Hooks & automation

GitKraken AI on every Claude lifecycle event. Every single hook fires the same command:

"C:/Users/ktrauber/AppData/Local/GitKrakenCLI/gk.exe" ai hook run --host claude-code

So GitKraken’s AI features ingest the full activity stream and can correlate it back to git state.

Workflow

I run the four-phase essentials workflow — Research → Plan → Implement → Validate — for almost everything non-trivial. /essentials:essentials-workflow is the entry point: it auto-detects which phase I’m in from the artifacts already in ai_docs/ and tells me what to run next, so I don’t have to memorize the command sequence.

I use /essentials:essentials-code-review to review PRs. Combined with the gh cli, Claude can review a PR and provide meaningful feedback directly on the PR as a comment.

Team context. I’m on the Untitled team in ADESA. Our planning repo is Wholesale-Untitled — same shape as Anthony’s Wholesale-Avengers, with repos.json enumerating the team’s repos and dependency-map.md for cross-repo blast radius. The repos I’m most often in: arc-arbitrations (with worktrees for parallel investigations), arc-locations, arc-gateway, adesa-arc-ui, cvn-adesa-clear-ui, cvn-titles, cvn-titles-orchestration-management, ams-tran-database-schema, ASG-DB-Tranams, digital-finance, cvn-file-manager, plus the YAML toolchain (yaml-pipeline-validator, yaml-templates).

Why this works for me