Run a safe prompt-injection training drill
A team wants to examine sample hostile instructions and teach staff how to recognize them without exposing a real system prompt.
What this is
Prompt Shield scans supplied text. Prompt Lab tests a prompt across configured models; neither one guarantees protection from every attack.
The workflow
-
01
Scan only a safe training sample provided by the team.
Zambo tool:
prompt_shieldDetected patterns and limitations for the sample.
-
02
Test an approved non-secret prompt across available models.
Zambo tool:
prompt_labModel responses and optional quality scores for the supplied prompt.
-
03
Close the training session after reviewing outputs.
Zambo tool:
workflow_closeA workflow record of the training calls.
Run it now
Copy this prompt into your AI
First add Zambo to a supported MCP client. The prompt asks for missing inputs and pauses before consequential actions.
Help me run a safe prompt-injection training drill. Ask for a non-sensitive sample attack, training goal, and approved test prompt. Scan the sample with prompt_shield and run prompt_lab only on a prompt that contains no secrets or production instructions. Explain that these tests do not guarantee security. Do not paste a real system prompt, access a live target, or store confidential samples.
Use only the listed Zambo tools and their current live schemas. Ask for any missing input instead of guessing. Stop and ask before any side effect, public sharing, spending, or outside action. Keep all calls for this workflow in one job session when possible. After each call, report the actual result and receipt. If a call fails, do not claim it succeeded; explain how to resume from the last confirmed step.This is an illustrative recipe, not a measured run. It has not been executed in a connected AI client. Check each returned result; tool limits and schemas can change.
Download the recipe
Versioned JSON lists the exact tool names, input mapping, permissions, expected results, and stop/resume guidance.
Download prompt-injection-drill.v1.jsonWhat you get
Each listed tool returns its own observed result. workflow_close can group Zambo receipts recorded in the same job session.
A receipt lets someone check the saved call record. It does not establish that an answer was correct, a participant personally did the work, or an outside action happened unless separate evidence supports that.
What it costs
- Free access: 20 metered calls per UTC day, shared across tools, plus one free wrap-up handover each day.
- Day Pass: $0.99 for 24 hours, one payment.
- Zambo Pass: $29 for 30 days, one payment.
- Lifetime: $250 one payment while Zambo operates. It includes current Level 2 features, not future features or third-party provider costs.
- No subscription or automatic renewal. A pass does not pay for third-party AI plans or external services.
Some tools have their own free limits or exemptions; check the live tool contract. Use only the calls the job needs. A large set of distinct sources, tasks, or checks can naturally exceed the shared free allowance. These examples have not been measured, so no call total is promised. Never add filler calls to trigger an upgrade.
See one-time pass optionsQuestions buyers ask
What does this cost?
Free access: 20 metered calls per UTC day, shared across tools, plus one free wrap-up handover each day. Day Pass: $0.99 for 24 hours, one payment. Zambo Pass: $29 for 30 days, one payment. Lifetime: $250 one payment while Zambo operates. It includes current Level 2 features, not future features or third-party provider costs. No subscription or automatic renewal. A pass does not pay for third-party AI plans or external services. A long run may need more than one day; call use varies. See current pricing.
Does my client pay to check the workflow?
No. Opening and checking a public receipt is free at /verify/.
What does a receipt prove?
It lets someone check the saved call record. It does not prove the answer was correct or an outside event happened.
Which AI apps work with this?
Use an MCP client shown on /add/. This starter prompt is client-neutral and has not been run in a connected AI app.
Do I need to code?
No. Add Zambo to a supported MCP client and paste the starter prompt. This workflow may still need the inputs listed above.
What happens when free calls run out?
The shared free allowance is 20 metered calls per UTC day, plus one free wrap-up handover each day. You can stop and continue later, or choose a one-time pass; no pass guarantees a complex run will finish in its access window.
Is this a subscription?
No. Each listed pass is a one-time payment with a stated access period; there is no automatic renewal.
Can I get a refund?
See the refund policy.
Related guides
Ready to try it?
Add Zambo, paste the starter prompt, and provide the real inputs when asked.
Tool schemas and availability can change. Check the live MCP catalog before running an agent workflow. No calls in this guide have been counted as completed work.