القائمة
Use when coordinating the seven-stage extraction pipeline. Tells the Orchestrator when to delegate, when to gate, when to escalate.
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 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.
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 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 | orchestration-playbook |
| description | Use when coordinating the seven-stage extraction pipeline. Tells the Orchestrator when to delegate, when to gate, when to escalate. |
| when_to_use | Triggered by the Orchestrator at every state transition. |
1. consent → 候选人同意 + 边界确认
2. intake → Intake Agent: resume → CandidateProfile + Hypotheses
3. interview → Interviewer Agent: 关键事件追问 (多轮)
4. evidence_mapping → Evidence Mapper: 把每轮对话变成 evidence_card
5. mechanism_naming → Mechanism Modeler: evidence_cards → mechanism_cards
6. skeptic_review → Skeptic Agent: 反驳与降级
7. publish → Report Composer: extraction_report
| 何时委派 | 何时不委派 |
|---|---|
| 任务需要并行处理多个独立证据面向 | schema 校验 (走 validator) |
| 任务需要独立上下文避免污染 | PII 检测 (走 validator) |
| 任务需要独立的反驳或独立命名 | 置信度计算 (走 confidence_scorer) |
| 任务输出可结构化为 schema | 引文回链检查 (走 citation_checker) |
high severity findings 必须有 repair_actions 处理| 失败 | 回退 |
|---|---|
| 充足性门: hypothesis 不足 | 触发 Intake 重新解析, 用更细颗粒度 |
| 证据门: incident 数不足 | 继续 interview, 把 hypothesis 优先级最高的换一个角度问 |
| 命名门: 机制全部只支撑 1 张证据 | 触发 Mechanism Modeler 重新聚类, 若仍不足则降到 hypothesis |
| 一致性门: Skeptic 无法消除高严重度问题 | 强制降级为 probable; 若 confidence < 0.4, 移到 needs_more_evidence |
| 发布门: PII 未清理 | redact 后重试, 不允许跳过 |
Orchestrator 在每个阶段切换时只输出: