界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
数据与研究 · SKILL.md

cyber-threat-hunting

cyber-threat-hunting

【数据与研究】Conducts hypothesis-driven, evidence-preserving threat hunts with bounded negative conclusions and telemetry-gap handling. Use for hunt hypotheses, scoped offline analysis, or analytic follow-up; not unscoped scanning, malware execution, or generic intelligence summaries.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: cyber-threat-hunting guardrail: true description: Conducts hypothesis-driven, evidence-preserving threat hunts with bounded negative conclusions and telemetry-gap handling. Use for hunt hypotheses, scoped offline analysis, or analytic follow-up; not unscoped scanning, malware execution, or generic intelligence summaries. metadata: triggers: files: [] keywords: - threat hunting - hunt hypothesis - hunting query - negative conclusion - suspicious activity hunt - telemetry gap

Cyber Threat Hunting

Priority: P0 (CRITICAL)

Preserve provenance and bound every conclusion to authorized evidence.

Structure

Input: falsifiable hypothesis, behavior, entities, scope, sources, window, expected signal, stop condition. Process: define positive/negative observations; normalize without overwriting originals; test narrowest evidence; correlate by entity/time; record gaps and alternatives. Output: status, cited evidence, bounded conclusion, limitations, owner.

Rules

  • Active collection or network operations require cyber-authorization, documented scope, and runtime-proven controls; unsupported or missing scope blocks execution.
  • Use controlled synthetic/offline fixtures. Preserve query/version, fixture provenance, retention, clock, identity, and source limitations.
  • Status is confirmed, suspected, false-positive, blocked, or not-tested; only cited evidence supports a status.
  • “No evidence found in the examined dataset/time window” is bounded negative evidence, never proof of no compromise.
  • Include engagement/scope reference; skill/version/source; observation time; status; evidence references; limitations; accountable owner. Reuse shared skill IDs.

Anti-Patterns

  • Never call an empty or incomplete dataset clean.
  • Never treat one ambiguous hit as confirmed compromise.
  • Never collect live telemetry before authorization or invent hunt metrics.

References (lazy, primary)

我的收藏