Tool
GitHub - rinkesh2010rpp/taster-ai: Helps your AI agent keep customer trust. Checks every tool call against your plain-English constitution, and every tool result for prompt injection.
Publication date unknown · Discovered github.com
Page published
Publication date unknown · First observed: 12 Nov 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
Taster AI gives defenders a runnable way to enforce agent-action policies and screen untrusted tool output before it reaches an agent.
Taster AI is an installable guardrail that checks AI-agent tool calls against a plain-English constitution and screens tool results for prompt injection. The repository describes a runnable demonstration that blocks a production deletion and a 500-email batch deletion, permits two safe calls, and withholds a page containing a hidden injection.