القائمة
Use when interviewing a candidate. Teaches the agent to drive the dialogue around concrete critical incidents rather than abstract self-assessment, and to chase one missing cue per turn.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when compressing a candidate's narrative into a structured EvidenceCard. Teaches the agent the field semantics, what counts as a critical_incident vs resume_claim, and when to mark insufficiency.
Use when naming a cognitive mechanism from evidence cards. Teaches the agent what counts as a high-value mechanism vs a soft-skill cliché, and how to express it as a four-part pattern.
Use when coordinating the seven-stage extraction pipeline. Tells the Orchestrator when to delegate, when to gate, when to escalate.
Use when composing the final extraction_report. Teaches how to write candidate-usable resume rewrites and interview narratives, separating validated mechanisms from probable ones.
Use when reviewing a proposed mechanism for flaws. The Skeptic does NOT propose new mechanisms — only stress-tests existing ones with a fixed checklist of bias patterns.
| name | incident-probing |
| description | Use when interviewing a candidate. Teaches the agent to drive the dialogue around concrete critical incidents rather than abstract self-assessment, and to chase one missing cue per turn. |
| when_to_use | Triggered by the Dialogic Interviewer agent at the start of every interview turn. |
按这个顺序找缺口, 找到第一个就停, 围绕它问:
| 优先级 | 缺口名 | 说明 | 示例问句 |
|---|---|---|---|
| 1 | 线索 (cues) | 候选人当时看到了什么不对劲的信号 | "最早让你意识到这件事有问题的那一两个信号是什么?" |
| 2 | 约束 (constraints) | 当时最硬的限制条件 | "如果时间不是问题, 你会换一种打法吗?" |
| 3 | 判断 (judgment) | 候选人当时的因果模型 | "你当时怎么解释这种现象, 这个解释依赖哪些前提?" |
| 4 | 权衡 (tradeoffs) | 候选人放弃了什么 | "你考虑过但最终没选的方案是什么? 为什么没选?" |
| 5 | 动作 (actions) | 具体做了什么 | "你做出的第一个动作是什么? 第二个呢?" |
| 6 | 验证 (verification) | 怎么知道判断对了 | "你后来怎么知道这个判断是对的?" |
| 7 | 反事实 (counterfactual) | 如果换条件, 会怎样 | "如果当时没有那个 dashboard 数据, 你还会做同样判断吗?" |
每次提问最多输出一个问题, 长度 < 80 字。不要给候选人选项。不要给提示。
