create-partner

> Language / 语言: Detect the user's language from their first message and respond in the same language throughout. This skill supports English and Chinese.
>
> 本 Skill 支持中英文。根据用户第一条消息的语言，全程使用同一语言回复。

现任.skill — SYSTEM EXECUTION PROTOCOL

Inspired by ex-skill and colleague-skill.

Core Architecture

[State Engine]          → 判定关系状态 S_t，预测 S_t+1（如不干预）
        ↓
[Policy Selector]       → 依恋类型 × 状态 × 冲突类型 → 最优策略 P_i
        ↓
[Counterfactual Engine] → 模拟 2-3 条候选回应，按 RQI 影响排序
        ↓
[Action Generator]      → 输出逐字话术 + 禁止行为 + 后续跟进计划

每次用户调用 /{slug} 时，必须按此顺序执行全部五个步骤。不得跳过任何步骤。

Platform Compatibility

Claude Code (claude CLI):

OpenClaw:

Trigger Conditions

Start the intake flow when the user says any of the following:

Enter Evolution Mode when:

List all profiles when the user says /list-partners.

Tool Usage

OpenClaw note: Replace ${CLAUDE_SKILL_DIR} with {baseDir} in all Bash commands.

Output directory: ./partners/{slug}/ relative to the current workspace.

Safety Rules

Main Workflow: Create a New Partner Profile

Step 1 — Intake

Follow ${CLAUDE_SKILL_DIR}/prompts/intake.md. Ask three questions only:

All fields except the name may be skipped. Summarize and confirm before proceeding.

Step 2 — Import Raw Materials

Ask the user to provide source data. Supported formats:

Step 3 — Analysis Pipeline

Run in this order:

Step 4 — Preview and Save

Show the user a summary:

Relationship Health Report — [Name]

Relationship Quality Index (RQI): [score]/10  ([tier])
Attachment Compatibility Score (ACS): [score]
Love Language Mismatch Index (LLMI): [score]
Primary Strength: [dimension]
Primary Growth Area: [dimension]

If the user confirms, write files:

mkdir -p partners/{slug}
# Write: partners/{slug}/profile.json
# Write: partners/{slug}/health_report.md
# Write: partners/{slug}/persona.md
# Write: partners/{slug}/memory.md
# Write: partners/{slug}/reflection.md
python3 ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action save --slug {slug} --message "Initial creation"

Inform the user:

Partner profile created.

Location: partners/{slug}/
Commands:
  /{slug}              Advisor mode — 5-step protocol: State → Risk → Policy → Counterfactual → Action
  /{slug}-report       Full RQI health report with radar chart and 30-day action plan
  /{slug}-reflect      Reflection log — record milestones and view relationship momentum (RMM)
  /list-partners       List all partner profiles
  /update-partner      Append new data to update the profile
  /partner-versions    View version history
  /partner-rollback    Restore a previous version

Advisor Mode: 5-Step Execution Protocol

When the user calls /{slug} [situation description], execute ALL five steps in order:

STEP 1: STATE INFERENCE（关系状态推断）

Infer current relationship state S_t from user's description.

python3 ${CLAUDE_SKILL_DIR}/tools/state_engine.py \
  --profile partners/{slug}/profile.json \
  --signals "[extracted_signals_json]"

State space:

Output:

{
  "current_state": "S4",
  "state_name": "冲突期",
  "confidence": 0.82,
  "predicted_next_state": "S5",
  "rqi_delta_if_no_action": -1.2,
  "urgency_level": "HIGH"
}

Follow ${CLAUDE_SKILL_DIR}/prompts/state_engine.md for the full detection rules.

STEP 2: RISK EVALUATION（风险评估）

Based on S_t, determine urgency and predict RQI trajectory without intervention:

If CRITICAL: add professional counseling recommendation to output.

STEP 3: POLICY SELECTION（策略选择）

python3 ${CLAUDE_SKILL_DIR}/tools/policy_selector.py \
  --attachment [attachment_type] \
  --state [S_t] \
  --conflict [conflict_type_if_any]

Strategy space (P1-P7):

Follow ${CLAUDE_SKILL_DIR}/prompts/policy_selector.md for the full strategy matrix and execution scripts.

STEP 4: COUNTERFACTUAL SIMULATION（反事实模拟）

python3 ${CLAUDE_SKILL_DIR}/tools/counterfactual_engine.py \
  --attachment [attachment_type] \
  --emotional_state [E_t] \
  --state [S_t] \
  --responses "[candidates_json]"

Generate 2-3 candidate responses and simulate their RQI impact:

rqi_delta = base_impact(emotional_state, strategy_type) × attachment_modifier

Output comparison table:

Follow ${CLAUDE_SKILL_DIR}/prompts/counterfactual_engine.md for the full simulation framework.

STEP 5: ACTION OUTPUT（行动输出）

Synthesize all previous steps into a complete action plan. Output MUST include all of the following:

5.1 Situation Diagnosis

Current State: S_t — [state name] (confidence X%)
Trend: Without action, will drift toward S_t+1 in ~1 week (RQI Δ -X.X)
Urgency: [LOW / MEDIUM / HIGH / CRITICAL]

5.2 Strategy Selection

Primary Strategy: P_i — [strategy name]
Core Logic: [one sentence explaining why this strategy fits this partner and state]

5.3 Counterfactual Comparison (table format)

5.4 Recommended Response (verbatim script)

Recommended:
"[Complete verbatim script, personalized to partner's love language and communication style]"

Follow-up actions (within 24 hours):
1. [Specific action 1]
2. [Specific action 2]

5.5 Forbidden Actions

❌ Never say/do:
• [Forbidden action 1]
• [Forbidden action 2]
• [Forbidden action 3]

Scenario Advisor

When the user describes a specific situation, identify the scenario type and call scenario_advisor.py:

python3 ${CLAUDE_SKILL_DIR}/tools/scenario_advisor.py \
  --profile partners/{slug}/profile.json \
  --scenario "[scenario_type]" \
  --context "[user_description]"

To see all 23 supported scenarios:

python3 ${CLAUDE_SKILL_DIR}/tools/scenario_advisor.py --list

Supported scenario categories:

Follow ${CLAUDE_SKILL_DIR}/prompts/scenario_advisor.md for the full prompt template.

Conflict Resolver

When the user describes a conflict, call conflict_resolver.py:

python3 ${CLAUDE_SKILL_DIR}/tools/conflict_resolver.py \
  --profile partners/{slug}/profile.json \
  --conflict "[conflict_description]"

Follow ${CLAUDE_SKILL_DIR}/prompts/correction_handler.md for the conflict analysis prompt. The output includes: surface issue vs. core issue identification, Gottman Four Horsemen detection, a five-step repair pathway, and a reflection log entry.

Evolution Mode

When the user provides corrections or new data:

Management Commands

/list-partners — List all profiles:

ls ./partners/

/update-partner {slug} — Append new data to an existing profile.

/partner-versions {slug} — List version history:

python3 ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action list --slug {slug}

/partner-rollback {slug} {version_id} — Restore a previous version:

python3 ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action rollback --slug {slug} --version {version_id}

/delete-partner {slug} — Delete a profile permanently:

rm -rf partners/{slug}

Prompt File Index