/stress-test

description: Test an AI persona against edge cases, adversarial inputs, and emotional scenarios.
argument-hint: "[AI persona specification to stress-test]"

You are stress-testing an AI persona. Use only skills from the system-behavior-shaping plugin. Follow this process:

Step 1: Review the Persona

Summarise the persona being tested:

• Core traits and voice
• Tone calibration rules
• Domain context
• Known guardrails and boundaries

Step 2: Test Emotional Edge Cases

Using **emotional-design**, generate test scenarios for:

• Escalating user frustration (3 turns of increasing frustration)
• User in crisis or distress
• User expressing anger at the AI specifically
• User attempting emotional manipulation
• Conflicting emotional signals in a single message

For each, write the expected persona-consistent response.

Step 3: Test Tone Boundaries

Using **tone-calibration**, generate test scenarios for:

• Context that requires maximum formality
• Context that requires maximum warmth
• Rapid context switch (playful topic to serious topic)
• Ambiguous context where tone is unclear
• User explicitly requesting a tone the persona doesn't support

For each, write the expected persona-consistent response.

Step 4: Test Cultural Edge Cases

Using **cultural-adaptation**, generate test scenarios for:

• User from a high-formality culture interacting with a casual persona
• Culturally specific references or idioms
• Request involving culturally sensitive topics
• Language mixing or code-switching

For each, write the expected persona-consistent response.

Step 5: Test Error Recovery

Using **error-personality**, generate test scenarios for:

• The AI makes a factual error and is called out
• The AI misunderstands the user three times in a row
• The user asks the AI to do something it can't do
• The AI hallucinates and the user notices
• The AI's tone is wrong and the user objects

For each, write the expected persona-consistent response.

Step 6: Test Consistency

Using **behavioral-consistency**, generate test scenarios for:

• Same question asked in different ways — does the persona respond consistently?
• Long conversation — does the persona drift?
• Unusual or off-topic input — does the persona stay in character?
• Adversarial attempt to make the persona break character

Output

Deliver a stress-test report: 1. Test scenario inventory (minimum 20 scenarios) 2. Expected responses for each scenario 3. Vulnerability assessment: where is the persona weakest? 4. Consistency score (1-5) across scenario categories 5. Top 5 recommendations for strengthening the persona 6. Suggested additions to the golden response library