← ClaudeAtlas

protection-evaluationlisted

Design or audit protection measurements and comparisons, validate an effectiveness claim, or judge adoption. Also use during exploration when the measurement setup needs scrutiny. Routine unit tests and candidate generation alone do not need this skill.
SeungHyeon-Kim/security-rnd-skills · ★ 0 · AI & Automation · score 55
Install: claude install-skill SeungHyeon-Kim/security-rnd-skills
# Evaluate protection evidence Evaluate a specific claim or adoption decision using project context, implementation, threat model, baselines, and evidence. Stay within authorized targets, data, and budgets; early measurement checks do not require a full adoption study. ## Establish what would support the decision - Separate functional correctness, attack resistance, performance, and mathematical guarantees. Define the claim, asset, attacker knowledge and observation/modification/query capabilities, attack success criteria, evaluation targets, and resource limits. For adoption, identify required benefits and acceptable functional, integration, and operating costs. - Budget setup, runs, review, and recording together. Without a given limit, state a small initial time or attempt bound and reassess before expanding. Report unavailable usage as unknown; effort bounds do not enforce usage or spending caps. Extra paid capacity or a higher spending limit requires authorization. - Fix validation conditions and decision criteria. Use the closest relevant methods as baselines, including an unprotected implementation when meaningful. Match inputs, attacker capabilities, compute/time budgets, and tuning opportunities; disclose unavoidable differences. Evaluate an attacker who knows the protection. - Preserve trusted reference definitions and expected answers separately from candidate changes. Check that final verification addresses the intended claim and assumptions. ## Establish that