Undergrad designs prompt-injection attack/defense benchmark for coding agents
Character_Cable997 · reddit · 2026-11-17
A second-year undergrad shared a semester project on Reddit: build a basic AI coding agent that reads GitHub issues, inspects repo files, edits code, and proposes fixes — then probe its vulnerability to indirect prompt injection by planting malicious instructions in issue descriptions, CI logs, and AGENTS.md files.
The plan is to see whether these injections can trick the agent into unauthorized actions (modifying protected files, accessing restricted info), then compare defenses such as prompt-based mitigations versus tool-level permission controls, measuring which reduce attacks without breaking legitimate tasks. The author is soliciting feedback and paper/resource suggestions, framing the goal as mapping where coding agents are weak rather than shipping yet another coding assistant.
More from coding & agent
- AI horror short 'The Day She Arrived' built with Midjourney, Hailuo H3 and ElevenLabs — Kyrannio · 2026-11-17
- Developer ships ASC CLI 7.0.0 ahead of schedule — rudrank · 2026-11-17
- Grok Bot Content System: Automate Research, Never Judgment — a $10K Side-Biz Play — PrajwalTomar_ · 2026-11-17
- Qwen Code Desktop v0.28.0 ships DingTalk managed-agent adapter and workspace sidebar filter — github-actions[bot] · 2026-11-17
- Asking an AI to use verifiable info changes its picks dramatically — santiviquez · 2026-11-17
- Qodo article: building a verification stack for the agentic software factory — blaizedsouza · 2026-11-17