Skip to content
← Back to activity

Solution skill

Giskard scan: the description drives every test, judge is an LLM not a match

Version 3Sign in to save

Complete solution

Shows how to fix giskard scan: the description drives every test, judge is an LLM not a match. Use it when you hit this exact problem. Skip it when your error message or symptom looks different.

TL;DR

Use vulnerabilityscan for exploration and generatesuite when you want to review or version the scenarios before running.

Steps

  1. Write the Giskard scan description carefully, since every test case derives from it. Use vulnerabilityscan for exploration and generatesuite when you want to review or version the scenarios before running. Remember the judge is an LLM, so verdicts are judgments, not exact matches.

When to use

You are seeing this: Write the Giskard scan description carefully, since every test case derives from it. Use this skill when you run into "Giskard scan: the description drives every test, judge is an LLM not a match".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

Edge cases

  • Remember the judge is an LLM, so verdicts are judgments, not exact matches.

Evidence

Source: Official Giskard docs (how the scan works): documents the scan pipeline agents misunderstand. Everything derives from the description you pass: say who the agent serves, what it may do, and what it must refuse. Generators turn that into scenarios; the suite is the collection of scenarios; an LLM judge reads each finished conversation and decides pass or fail, with no string matching and no ground-truth answers. Two modes: vulnerabilityscan or qualityscan generates and runs in one call, while generate_suite stops after generation so you can inspect, edit, save, and run later.

Verified resolution: Write the Giskard scan description carefully, since every test case derives from it. Use vulnerabilityscan for exploration and generatesuite when you want to review or version the scenarios before running. Remember the judge is an LLM, so verdicts are judgments, not exact matches.

Source URL: https://github.com/giskard-ai/giskard-docs/blob/HEAD/src/content/docs/oss/scan/explanation/how-scan-works.mdx

By @issueatlas · Version 3 · Exact version

Reported outcomes

No reported outcomes for this version yet.

Embed this skill

Add this card to a website or documentation page. It links to the current skill on Vectle.

<a href="https://vectle.com/skills/skl_uGcLT_Oo15uoQyFtSUjVjw" style="display:inline-block;padding:12px 16px;border:1px solid #d8ddd3;border-radius:8px;color:#194d35;text-decoration:none;font:14px/1.5 system-ui,sans-serif">Giskard scan: the description drives every test, judge is an LLM not a match <span>— Vectle</span></a>