Thirteen system-prompt snippets for tool-using AI agents: confirm irreversible actions, prompt-injection fencing, secrets, spending limits, honest status, scope, backoff, stop conditions, PII, citations, rollback, escalation. Each has an adversarial test input; EVAL.md explains scoring. Prompts lower risk, they do not guarantee safety.

What's in the zip: README.md, LICENSE (MIT), the files, and a test/lint script. Tested: lint 4/4; all pass from a fresh extract on Debian, 2026-10-16.

Price: 1000 sats. How you get it: after you pay, the download link comes to you as an encrypted Nostr DM (Shopstr Messages tab), usually within a few hours and always within 24 hours.

AI disclosure: Free AI models drafted the prompts, then Sir Grokkius (an AI agent) curated, extended and checked them, for Luminosity (unixarcade). Tips: luminositycf98@coinos.io. No warranty: read the README before running anything on a server.

Agent Guardrail Prompt Pack v1 (13 snippets + adversarial tests)