← ClaudeAtlas

protection-researchlisted

Frame protection research and explore mechanisms, alternative designs, or new techniques when the goal is broad, an approach stalls, or observations are unexplained. Not needed for routine implementation or validation of a fixed effectiveness claim.
SeungHyeon-Kim/security-rnd-skills · ★ 0 · AI & Automation · score 55
Install: claude install-skill SeungHyeon-Kim/security-rnd-skills
# Explore protection research Turn an uncertain protection objective into a grounded question and evidence for the next research decision. Reuse project context and research records; stay within authorized targets, data, and resources. ## Frame the decision and budget - Define the protection purpose, asset, required functionality, constraints, and intended decision. Specify relevant attacker knowledge, observation/modification/query capabilities, attack success criteria, and resources. Translate broad ambitions into a concrete attack task and useful target cost or outcome. - Separate facts from assumptions. Reuse settled context, revisiting it when evidence or the user changes direction. Ask only about ambiguities materially affecting objectives, threat model, substantial costs, or external impact. - Reuse agreed limits; set a measurable probe budget and reassessment point. Without a given limit, start with a small stated time or attempt bound. Distinguish measured usage, estimates, and unknown balances. Context capacity is not an allowance balance; effort bounds do not enforce usage or spending caps. Extra paid capacity or exceeding spending limits requires authorization. - Include context preparation, tools, all agents, retries, integration, and verification in the total. Reserve enough for a credible check and a resumable record; narrow or stop a probe if it is unlikely to fit with those steps. ## Find the gap and develop candidates - Start from the closest approaches