VexorGate Open-Sources an Agent Guardrail Catching Out-of-Scope Tool Calls
Ai-test-5664 · reddit · 2026-11-21
The author open-sourced VexorGate (v0.4), a checker that vets each agent tool call against what the user actually asked, returning allow / ask / block with a plain-language reason. It targets valid-but-overreaching calls: one refund requested, two issued; staging named, production restarted; "share with the team" turned into "anyone with the link."
How it works
- Plain-code rules check hard facts first: environment, count, amount, audience, repeated calls
- An optional small decision model handles what rules can't
- Runs rules-only with no API key, fully local; works over existing agent logs and as a Claude Code hook
Early results (small synthetic set): caught 16/18 risky calls with 2 false positives out of 19 safe ones; rules alone caught only 3/18.
The author seeks real agent traces, false-block cases, and volunteers writing 5–10 private test cases (only scores are shared).
More from coding & agent
- Mnemos 2.0: agent memory designed to preserve what matters for the next session — RileyRalmuto · 2026-11-21
- InferenceBench ranks agents on LLM serving speedup: Claude Opus 5.5 tops at 12.08x, agents lose to simple hyperparameter search — maksym_andr · 2026-11-21
- 4 architecture patterns to tame multi-agent system reliability, from a 27-year reliability engineer — bibryam · 2026-11-21
- AutoCut + FrameAtlas: open-source local-first video editing driven by Claude Code via MCP — moheshmohan · 2026-11-21
- InfoQ: What an agent harness is, and two ways to build one — bibryam · 2026-11-21
- The hidden 70 minutes: a Reddit cost breakdown shows agent workflows can cost more human time than manual — AlarmingPepper9193 · 2026-11-21