AgentSkillsCN

freud-detection-ai

为游戏运营设计并验证AI驱动的异常/欺诈检测工作流。在定义信号特征、模型评分阈值、调查路径，或验证检测管道的可靠性和假阳性控制时使用此技能。

SKILL.md
--- frontmatter
name: freud-detection-ai
description: Design and validate AI-driven anomaly/fraud-style detection workflows for game operations. Use when defining signal features, model scoring thresholds, investigation routing, or validating detection pipeline reliability and false-positive controls.

Freud Detection AI

Use this skill to implement anomaly-detection workflows with explainable gating and review paths.

Workflow

  1. •Define scope and constraints.
  • •Define detection scope, signals, threshold policy, and review SLA.
  • •Capture objective metrics, bounds, and release blockers.
  1. •Design implementation plan.
  • •Design model-scoring flow, fallback heuristics, and escalation rules.
  • •Keep ownership and dependency boundaries explicit.
  1. •Execute and iterate.
  • •Implement in small, traceable increments.
  • •Record run/build context for reproducibility.
  1. •Validate contract integrity.
  • •Validate threshold outcomes, alert quality, and investigation traceability.
  • •Treat contract breaches as blockers.
  1. •Prepare handoff.
  • •Deliver detector configuration diff, alert routing updates, and runbook.
  • •Include exact commands and acceptance criteria.

Output Contract

Return:

  1. •Context: goals, assumptions, constraints.
  2. •Validation: pass/fail checks and key deltas.
  3. •Changes: concrete file-level updates.
  4. •Commands: commands and expected outputs.
  5. •Risks: unresolved issues and limits.

References

  • •references/workflow.md: detailed execution flow.
  • •references/checklist.md: sign-off checklist.

Execution Rules

  • •Keep decisions measurable and reversible.
  • •Keep validation criteria explicit before iteration.
  • •Escalate unbounded false-positive risk and opaque scoring logic as blockers.