Pricing
The unit is one full check: a model-drafted test suite, a run on real models with and without the skill, an LLM judge, an HTML report. The quick scan is always free and unlimited.
- Credits never expire
- Check history in your account
- Credit returned if the run did not happen
- Credits never expire
- Check history in your account
- Credit returned if the run did not happen
- Credits never expire
- Check history in your account
- Credit returned if the run did not happen
Request a pack
Online payment is not connected yet. Leave a contact and we will reply with an invoice or add the credits by hand.
FAQ
What exactly happens in a full check?
If the skill ships no evals/spec, a model drafts 4–6 cases, trigger phrases and assertions (the draft is shown in the report). Each case then runs twice: with the skill in the system prompt and without. Answers are checked by deterministic assertions and an LLM judge. Separately we test whether the skill triggers on the right phrases among eight decoy skills. Result: ACTIVE / DEGRADED / OBSOLETE, pass rate, uplift vs. baseline and an HTML report.
Which models are used?
Several models from different families: some run the cases, a separate one judges the answers. We do not publish the line-up and change it as stronger models appear. The skilltest CLI runs the same tests on any models, including GigaChat and YandexGPT through an OpenAI-compatible gateway.
What happens to my files?
Check files are kept for 30 days and then deleted; the result stays at its secret link. Scripts are never executed; secrets found are masked in the report. Skills added to the rating are public.
The check failed, was the credit charged?
If the model run did not happen (guard blocked the skill, provider error, budget), the credit or free attempt is returned automatically.