Zynkr Recruiter

---
name: zynkr-recruiter
sheetId: "7.07"
description: "End-to-end recruiting pipeline — JD authoring, resume screening, interview design with scorecard, and calibration / final decision — bilingual (Chinese-English) outputs throughout."
category: people-talent
project: zynkr-recruiter
platform: claude
status: Done
author: Jane Liao
input: "A hiring need from a hiring manager — anywhere from rough notes to a full JD + resume stack"
process: "Stage 1 JD architecture → Stage 2 resume screening with hybrid weighted scoring → Stage 3 competency-based interview design + scorecard → Stage 4 calibration & final decision recording"
output: "JD draft (YAML), candidate fit scores with HR Decision Summary, bilingual interview kit + scorecard, and a final decision sheet with bilingual candidate communication templates"
synergy: []
type: agent
skills: ["recruiter-jd-architect", "recruiter-resume-screener", "recruiter-interview-question", "recruiter-calibrator"]
house-style: bound

---

# Recruiting Pipeline

```bash
npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill zynkr-recruiter
```

Run the full hiring loop — from rough role notes to a calibrated hire/no-hire decision — through four specialized agents. Use this skill at any entry point in the loop; the orchestrator will route to the right agent and pass artifacts forward.

---

## Step 1 — Identify entry point

Ask the user:

```
Where are you in the hiring process?
1. Starting from scratch — only have rough role notes
2. Have a JD — ready to screen candidates
3. Have shortlisted candidates — ready to design interviews
4. Interviews done — ready to calibrate and decide
```

Use the answer to determine the starting stage. Skip earlier stages if their artifacts are already provided.

---

## Step 2 — Stage 1: JD Architecture (subagent)

Display:

```
---------------------------------------------
Stage 1: JD Architecture (The RoleCrafter)
---------------------------------------------
```

Launch the `recruiter-jd-architect` agent (`./agents/recruiter-jd-architect.md`) using the Agent tool.

The agent will collect: role context → daily tasks → success metrics → requirements → reporting line → career level → work mode → optional salary → optional stakeholder alignment.

Output a structured JD in YAML with version label. Store as `JD_BLUEPRINT`.

---

## Step 3 — Stage 2: Resume Screening (subagent)

Display:

```
---------------------------------------------
Stage 2: Resume Screening (Hybrid Weighted Scoring)
---------------------------------------------
```

Ask the user to provide candidate resumes (text or file). Optionally collect 2–3 sentences about company culture and a target shortlist size.

Launch the `recruiter-resume-screener` agent (`./agents/recruiter-resume-screener.md`), passing `JD_BLUEPRINT` and the resumes.

The agent runs three layers of evaluation:
1. **Hard Skill Gate** — caps Relevance if must-have skills missing
2. **Domain Fit Check** — caps Relevance for cross-function mismatches
3. **Hybrid Weighted Scoring** — Relevance / Value / Behavior / Career / Brand with adaptive weights

Threshold = 60. Output: ranked candidates with HR Decision Summary per candidate. Store as `CANDIDATE_PROFILES`.

---

## Step 4 — Stage 3: Interview Question Design + Scorecard (subagent)

Display:

```
---------------------------------------------
Stage 3: Interview Design (Competency + Scorecard)
---------------------------------------------
```

For each shortlisted candidate, launch the `recruiter-interview-question` agent (`./agents/recruiter-interview-question.md`), passing `JD_BLUEPRINT` and the candidate's `Candidate Strength Profile` from Stage 2.

The agent produces:
1. Competency model (5–7 items)
2. Interview round assignment
3. Competency-based question bank (behavioral / situational / technical / culture) with probes and 1–5 anchors
4. Candidate-specific risk questions (6-dimension risk model)
5. Weighted scoring rubric with behavioral anchors
6. Bilingual scorecard ready for HR

Store as `INTERVIEW_KIT_<candidate>`.

**Important:** This agent does NOT score candidates. Scoring happens after interviews, in Stage 4.

---

## Step 5 — Stage 4: Calibration & Decision (subagent)

Display:

```
---------------------------------------------
Stage 4: Calibration & Final Decision
---------------------------------------------
```

After interviews complete, ask the user to provide:
- Interview transcripts (if available)
- Each interviewer's independent scores and notes
- Whether AI should extract evidence only, or also draft calibration summaries

Launch the `recruiter-calibrator` agent (`./agents/recruiter-calibrator.md`), passing `JD_BLUEPRINT`, `INTERVIEW_KIT_<candidate>`, and the interview artifacts.

The agent produces:
1. Behavioral evidence per competency (extracted from transcripts)
2. Pre-calibration review (score discrepancies, contradictions)
3. Calibration meeting discussion guide
4. Final decision sheet (Hire / No Hire / Hold with rationale, risks, mitigations)
5. Bilingual communication templates (internal to recruiter + external to candidate)

Store as `FINAL_DECISION_<candidate>`.

---

## Step 6 — Hand-off and follow-up

Compile per-candidate deliverables:
- JD Blueprint (Stage 1)
- Resume Screening summary (Stage 2)
- Interview kit + scorecard (Stage 3)
- Final decision sheet (Stage 4)

Ask:
```
Next steps:
1. Send candidate communications (internal + external templates)
2. Move shortlisted candidates to onboarding handoff
3. Iterate on a specific stage (e.g., refresh JD, add candidates, redo calibration)
```

---

## Rules

- All deliverables must be bilingual (Chinese-English, side by side)
- Never invent candidate evidence not in the transcript
- Never modify the original scorecard during calibration
- Stage 3 designs questions only; Stage 4 evaluates — keep them separate
- Threshold for shortlist is 60; below threshold, suggest contract or alternative roles instead of forcing hire

## House style

Writing style is **not owned by this file**. The house voice lives in two Google Docs under
`[@] 寫作指南` (`12DBdFz3SK22ie9im_ThFMI7IBRXsTZsV`), read at runtime:

- 《[2.0] Zynkr 通用風格指南 House Voice》 `10bOIQwRm9Pxwgct4hlwCwK_B4Pipai1HqBPZKzyRHSE` —
  the universal core, plus the addendum for this surface
- 《[3.2] 禁用詞清單 Forbidden Words》 `1N5sHLP4qzmmhpCGsi6KElxi1z0MFe4QZ0Q_35T10Uyg`

Read both before producing client- or reader-facing text, and scan the draft against 《[3.2]》
before handing it over. If Drive is unreachable, say so in the output rather than proceeding
unchecked. Never re-implement either list inside this file.