Blindspot: Autonomous Local AI Code Reviewer

Problem: Standard Git diffs miss deep architectural flaws, unhandled side-effects, and legacy anti-patterns. While automated AI code reviewers exist, most operate as monolithic cloud bots that dump entire diffs into a single generic prompt. This approach leads to hallucinated fixes, missed edge cases, and high latency. Furthermore, existing tools force developers into a tedious context-switching loop: commit code, push to remote, wait for CI/CD, read comments on GitHub, and manually copy-paste fixes back into their editor—all while burning through expensive cloud API credits on every draft commit. Solution: I built Blindspot as an autonomous, dual-engine code governance agent designed to secure enterprise codebases without slowing developers down. Blindspot bridges the gap between local terminal velocity and cloud CI/CD governance through a two-phased workflow: Enterprise CI/CD Integration (The Cloud): When a pull request is opened on GitHub, Blindspot leverages IBM Bob 2.0 to act as an automated gatekeeper. Instead of relying on a single prompt, Blindspot orchestrates three specialized subagents to run in parallel: Supply Chain Agent: Hunts for dependency typosquatting and unpinned floating version updates. Architecture & Auth Agent: Cross-references internal guidelines to trace route changes, ensuring proper authentication middleware and catching dangerous execution side effects. Legacy Modernization Agent: Scans for deprecated Node.js APIs (e.g., new Buffer()) and ES5 anti-patterns. A Manager Agent synthesizes these findings into a unified, actionable Markdown report posted directly to the PR, utilizing native GitHub suggestion blocks for 1-click visual remediation. Developer Terminal Integration (Local): Developers shouldn't have to wait for a remote CI/CD run to check their code. I built an interactive, local CLI for Blindspot so developers can audit uncommitted changes directly in their terminal using cost-free stubs. The CLI identifies vulnerabilities and provides an interactive [y/N] prompt to instantly inject secure refactors into the local file system. What Makes Blindspot Different from Existing AI Code Reviewers: Multi-Agent Parallel Orchestration vs. Monolithic Prompts: Rather than sending a massive diff to a single general-purpose prompt, Blindspot splits review responsibilities across domain-specific subagents equipped with dedicated repository tool-calling capabilities (readFile, readPackageJson). Active Local Remediation vs. Read-Only Cloud Comments: Existing tools only tell you what's wrong on a web interface. Blindspot actively fixes the issue on your local disk with a single keystroke (y), instantly updating the file so you can inspect the visual red/green diff directly in your IDE before ever pushing code. Cost & Velocity Awareness: Blindspot prevents API token exhaustion by providing zero-cost local terminal verification for rapid daily development, reserving full-scale multi-agent reasoning for official CI/CD pull request gates. Blindspot transforms code review from a reactive, manual bottleneck into a proactive, autonomous security layer.