Bob X-Ray — Legacy Knowledge Rescue

THE PROBLEM IBM has 80,000+ developers and decades of legacy code. A new joiner takes 3-4 weeks just to understand what a codebase does, because architecture docs are missing or outdated. Senior engineers burn 30-40% of their time answering the same onboarding questions. The most critical knowledge — pricing rules, auth quirks, deprecated contracts — lives only in tribal memory. Bus Factor = 1. THE SOLUTION Bob X-Ray — Legacy Knowledge Rescue. Give Bob a repository, and it generates a complete Onboarding Kit for new developers: 1) ONBOARDING.md — what the project is, 5-minute setup, folder guide, and the "known quirks" nobody documents (the traps that break production) 2) ARCHITECTURE.md — Mermaid diagrams reverse-engineered from code: system map, module dependency graph (nodes color-coded by risk), request sequence diagram, ER diagram, and the order status state machine 3) RISK-REPORT.md — file-by-file risk scores (1-10), a Bus Factor report (flagged: 1), top knowledge landmines, and prioritized quick wins 4) STARTER-TASKS.md — 3 safe, reviewable first tasks for the new developer, each with steps, estimates, and TRAP warnings about legacy gotchas Plus a self-contained dark dashboard (index.html) presenting the whole kit with rendered diagrams — works offline, mermaid library embedded. HOW WE USED IBM BOB (all 4 capabilities) - ASK mode: full repo analysis — tech stack, entry point, boot flow, module dependency map, and complexity ranking explained like a new joiner briefing - AGENT mode: created the 14-file legacy demo repo; generated all 4 kit documents; built the dashboard — with permission-gated file operations - ORCHESTRATOR / SUBAGENTS: the Onboarding Kit was produced as 4 parallel subtasks ("All tasks completed! 4/4") — docs, diagrams, risk report, tasks - DOCUMENT UNDERSTANDING: Bob cross-checked the repo's outdated 2019 API doc against the actual code and produced a "Docs vs Code" lie table — 8 confirmed conflicts: wrong auth scheme (Basic vs Token), wrong token TTL (24h vs 72h), "no coupon support" (two coupons fully implemented), "no refund endpoint" (one exists), hard-delete claim (actually soft-delete), wrong /v1/ URL prefix, missing R (refunded) status, undocumented /health + /status. PROOF IT'S REAL The generated legacy repo actually runs: `npm start` → port 4790 (hard-coded because an external finance cron job depends on it), `curl /health` → 200 OK. Demo repo: oms-api — a Node.js/Express/PostgreSQL order-management service with a tiered pricing engine, coupon system, token auth, and a FAKE_MODE fallback that keeps endpoints alive when the database is down. IMPACT - Onboarding: 3-4 weeks → Day 1 productive - Senior mentoring load: -30-40% (repetitive questions answered by the kit) - Knowledge risk: Bus Factor made visible and actionable, landmines documented - Fully repeatable: one command re-runs the whole analysis on any repo