What problem does it solve? Teams often ship systems without knowing where they break under stress, malformed input, or adversarial attack. This Skill runs a bounded adversarial testing workflow that maps the attack surface, locks breakage thresholds before attacking, and produces confirmed failure signals instead of vague suspicions. ## Core Features & Use Cases - Attack surface mapping and threat modeling: Enumerates entry points, trust boundaries, resource limits, and agent tool/permission surfaces before any probe runs. - Bounded attack loop with locked thresholds: Runs load, soak, spike, chaos, fault injection, fuzzing, concurrency, and prompt-injection probes against pre-defined SLOs and safety boundaries, recording a tried/broke/held ledger. - Safety gates and handoff: Requires human approval and blast-radius signoff before live or destructive probes, and routes confirmed breakages to debug-investigation rather than diagnosing root cause. - Use Case: Before launching a new API, run a default-depth probe that fuzzes endpoints, injects latency into dependencies, and verifies rate-limit SLOs hold, then hand any confirmed breakage to debugging with evidence attached. ## Quick Start Ask the agent to run an adversarial probe against your staging service or agent workflow, specifying the target environment, allowed attack modes, and the SLOs that must hold.
