Guides
Practical how-to guides.
- How to test a chatbot before launch Pressure-test customer-facing chat before users teach it new tricks.
- How to pressure-test a RAG app Your sources are not enough if the AI follows the wrong instructions.
- How to test AI agents that can take actions The risky part is not what the agent says. It is what it can do next.
- How to choose attack packs for your AI product Match packs to target type, actions, retrieval, and release importance.
- How to read a Release Gate Turn scan output into a meeting-ready ship recommendation.
- How to fix prompt injection findings Reduce instruction hijacking without pretending the problem disappears forever.
- How to retest an AI feature after a fix Prove the break is gone before you call it resolved.
- How to create a release policy for AI features Encode what should block release for your team and product stage.
- How to share an AI release decision with your team Use release reports to align eng, product, and leadership on one snapshot.
- Why happy-path testing misses AI failures Your demo script is not the conversation your users will have.
- What to test before launching an AI support bot Support bots face public pressure, policy questions, and injection attempts daily.
- How to tell whether an AI finding should block release Severity, repeatability, audience, and actions matter more than fear.
- Connect Agnostics MCP to your coding agent Wire the Agnostics MCP server into Cursor or Claude Desktop for local code scans, scan uploads, and Release Gate checks.