## Problem Only ~22%% of the catalog addresses AI/LLM-specific risks the actual core of vibe coding. ## New entries (each anchored to external evidence) - [ ] Prompt injection in agent loops OWASP Top 10 for Agents 2026 - [ ] Over-mocked / assertion-free tests arxiv:2602.00409 - [ ] Context window gaming / context poisoning HumanLayer ACE - [ ] Hallucinated imports / phantom deps arxiv:2603.28592 - [ ] Echo-chamber review (same model generates & reviews) - [ ] Architecture drift in long sessions - [ ] Tokenomics debt (vexp.dev: $300-450 vs $90-135/mo) - [ ] Spec ambiguity cascade **Target:** AI coverage 22%% !92 e45%%.
Problem
Only ~22%% of the catalog addresses AI/LLM-specific risks � the actual core of vibe coding.
New entries (each anchored to external evidence)
Target: AI coverage 22%% !92 e45%%.