ModelXGit: AI Repo Onboarding with IBM Bob

Problem: When developers join a team or pick up an unfamiliar project, onboarding is slow and risky. Setup instructions are often outdated, the architecture lives in senior developers' heads, and a frontend developer and a database developer get the same wall of files with no starting point for their role. Worse, secrets that were "deleted" months ago still sit in old git commits, and nobody checks. Time, effort and errors are all too high. Solution: RepoPilot is an AI onboarding system powered by IBM Bob. The developer pastes a GitHub or GitLab link and confirms. RepoPilot clones the repo with its full history and runs eight Bob skills as a pipeline: - Automatically, in parallel: setup-dependencies (install plan, .env and API keys, verified run commands), tech-stack-detection (every item backed by a file), git-history-secret-audit (every branch and tag), secret-precommit-scanner (plus a hook that blocks leaking commits), then architecture-diagram once the stack is known (Mermaid plus an interactive graph with drill-down to files). - On demand: codebase-qa, a role-based Q&A. Say "I'm a backend developer" and Bob gives a briefing for that role, answers with real file and line references, admits when something doesn't exist, and suggests 2-3 follow-up questions. And readme-generator, which creates or completes the README from verified facts. Finally, one click opens the repo in Bob IDE with the same skills and a Codebase Q&A mode installed, so the developer keeps working with Bob. Target users and interaction: New hires, interns, open-source contributors, and team leads who onboard people. They use a web dashboard: paste a link, watch live progress per skill, retry failed steps, explore the architecture graph, read the reports, ask questions, and download everything as one onboarding pack. What makes it creative and unique: - A real multi-step agentic workflow, not "AI helps me write code": skills declare their own triggers, dependencies and parallel groups, and the orchestrator runs them. - Role-based onboarding instead of a generic code explainer. - Security is built in, not bolted on: during pipeline runs Bob cannot execute commands, may only write to onboarding/ and README.md (everything else is reverted), and cannot see secret files because of a .bobignore we write into every clone. Every output is scanned and secret-looking values are masked. Secret scanners run locally, so raw secrets never reach the model. - Measured, not claimed: the backend records time, files produced, findings and Bob usage per run. Results: On a real IBM Bob run on our own repository, the automatic pipeline finished in 4 min 38 s. All 8 of 8 skills were completed, producing 15 files that all passed our output-contract checks, with 122 Bob tool calls and zero changes outside the allowed folders. A new developer goes from an unknown repo to setup steps, stack, architecture, a security audit, and answers for their role in minutes instead of days of reading code and asking around.