BSides Vancouver 2026

"Do you trust the authors of the files in this folder?" It's a prompt modern IDEs throw at developers, and most click past it by reflex. But as vibe coding, AI-assisted tooling, and automated agents accelerate software development, that implicit trust in established tools like VS Code and authoritative sources like GitHub has become a critical, highly exploitable attack surface - especially for non-technical vibe coders.

This 4 hour hands-on workshop places attendees directly in the mindset of an attacker targeting modern development environments. We move beyond traditional social engineering and focus on how trust is abused through the tools developers rely on every day: IDEs, agent harnesses, and package managers. After dissecting recent real-world cases of developer-targeted attacks and AI-agent vulnerabilities, we transition into labs where participants build and execute their own PoCs - including constructing malicious repositories from scratch to trigger invisible code execution on open, weaponizing hidden prompt injections to drive AI agents into running attacker-controlled commands, and standing up a custom "Claude Code"- style agent harness to attack ourselves.

Agenda:
⚠️ How your trusted IDEs would betray you by executing malicious commands automatically (spoiler - they didn't!)
⚠️ How agent harnesses like Claude Code and Gemini CLI can be abused across multiple trust boundaries
⚠️ How package managers like npm turn seemingly harmless actions, like a routine package update, into full compromise

Bring a laptop - we'll get our hands dirty and hack ourselves together.