How to use the 狗头军师恋爱顾问 agent skill (goutoujunshi)

What the goutoujunshi skill makes the agent do

The skill is written in Chinese and covers crushes, ambiguity, dating, relationship confirmation, choosing between several people, conflict, coldness, unequal effort, breakups, getting back together, infidelity, and marriage or family problems.

Its core order is to acknowledge the feeling first, separate facts from guesses, then give options you can carry out. For a message question it gives a sendable reply, timing, main risks and likely follow-ups. It reads chat screenshots or logs as evidence of what was written, not as proof of what someone secretly feels.

• Emotional grounding: names the feeling and the trigger before advice.
• Fact split: known facts, reasonable guesses and key unknowns are kept apart.
• Clear recommendation: one preferred option with reasons, plus up to three wording versions when there is a real trade-off.
• Next action: one small step, an observation window or a stop condition.

Install the skill from a ZIP file

Download the ZIP from the page where you found this tutorial, then unzip it. The unzipped folder should contain SKILL.md along with the other project files.

For a personal install, place the folder inside ~/.claude/skills/. For a single project, place it inside that project's .claude/skills/ folder. Use whichever matches how you work with the agent.

In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.

• Personal use across projects: ~/.claude/skills/
• One project only: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

You do not need a special command. Describe the situation in ordinary language and mention the relationship topic. The skill is meant to activate for crushes, dating, screenshots, breakups, conflict and similar situations.

On the first use, the skill asks a short intake: your MBTI type, a self-rated overall score from 0 to 100, your main strengths and weaknesses, and for each other person a codename, MBTI, score and current relationship. You can leave anything unknown blank or just tell the story. Each person gets a separate profile.

• Ask for a wording: "She replied with 哈哈 after I asked about Saturday. How should I follow up?"
• Ask for a decision: "I'm unsure whether to keep pursuing or step back. Help me compare."
• Share a screenshot or pasted log and say who is who, so the skill can map the speakers correctly.

Examples of useful requests

Short, specific requests get the most useful answers. If you only ask "how do I reply", the skill may need more context before it can judge timing and risk.

When you paste a long chat, the skill first locks the mapping between you and the other person. It prefers sender names or confirmed identities over guesses about which bubble is on which side, and it asks if the mapping is unclear.

• Message help: paste the last few messages and state your goal (invite, reply, clarify).
• Screenshot review: note which parts are visible text, which are your summary, and what you do not know.
• Repair after conflict: describe what happened in order and what you want the outcome to be.
• Choosing between two people: give the intake for each, and the skill compares them separately before ranking.

What it needs and how it handles data

The source lists no required dependencies. It does not run scripts and does not use the network. Its long-term memory is optional: it only works across tasks after you explicitly agree, and you can review, pause, revoke or delete what it stored.

For ChatLab data, the skill can only analyse files you have already provided or imported. It does not claim to read, decrypt or export records from chat apps such as WeChat or QQ.

• Requirements: none listed in the source.
• Scripts: none run.
• Network: none used.
• Memory: off unless you agree; it keeps limited fields and key events, not whole chat logs.

Safety limits and when to stop

The skill is built to refuse manipulative goals such as controlling, testing or pressuring another person. It will not help with coercion, stalking, spying, threats or fraud, and it will not promise that someone will fall in love with you.

It does not diagnose mental illness. If someone is in danger, the skill tells you to confirm your immediate safety and contact trusted support or local emergency services. An agent chat is not a substitute for police, hotlines or crisis care.

If someone clearly says they are not interested or asks you not to contact them, the skill advises stopping. A refusal of one specific time or method is not automatically read as rejecting the whole relationship.

• Emergencies, abuse, stalking, coercion or self-harm: contact local emergency or support services first.
• Refusals and requests for no contact: stop pursuing.
• You keep the final decision; the skill states key uncertainties.

Tips for getting better results

Give the skill the actual words used, the order of events and the time gaps. Treat the output as one perspective to check against what you know about the other person. If a reply in the output does not fit your style, say so and ask for a version in your own voice.

Keep each message to one main action: a reply, an invitation or a clarification, not all three together. The source suggests this so the other person can respond clearly.

• Paste exact text rather than summaries when wording matters.
• Correct the skill when it gets a profile detail wrong.
• Ask for follow-up branches: positive, unclear and no reply.