KRT Real-World Edge Testing

Turn “test the edge cases” into a reproducible failure campaign that exercises the real system and leaves it recoverable. Produce executable cases and evidence, not a generic checklist.

Load References

• Load references/safety.md before designing or running a campaign.
• Load references/campaign-contract.md when defining cases, fixtures, profiles, evidence, state, or completion criteria.
• Load references/failure-taxonomy.md while discovering and prioritizing failure modes. Select relevant categories; do not turn the taxonomy into mandatory busywork.
• Load references/evaluation-pack.md only when evaluating this skill's routing or behavior.

Choose A Mode

• Design: inspect the repository and produce the campaign specification without changing or exercising the system.
• Build: create fixtures, manifests, validators, runners, profiles, and runbook material; do not perform external operations.
• Safe execution: run offline and non-destructive cases after the applicable preflight checks pass.
• Stateful execution: run reversible mutations only after proving target isolation, recording affected resources, and validating recovery.
• Chaos: inject dependency or infrastructure failures only in a verified disposable environment and with explicit authorization for the exact affected target.