mindreader

🔮 mindreader-skill

Stop guessing what they really meant. Stop rehearsing in your head. Talk to their shadow first.

中文版 README · Demo · Quickstart · FAQ

⚠️ This is an early v0.2 project. Things will still break. The ethics are unresolved on purpose. Read the Why this exists section before starring.

---

Build a digital shadow of anyone real — your ex, your boss, the person you can't read, a difficult client, your parents.

Then do three things with it:

🕯️ Séance Mode /mindreader seance — Pull up an old conversation. Ask the shadow: "What were they actually thinking?"

🎭 Rehearsal Mode /mindreader rehearse — Before you say it to the real person, run it past their shadow first.

📐 Calibrate Mode /mindreader calibrate — Feed in a real conversation. The shadow predicts the second half. See how close it gets.

This is a hypothesis generator, not real mind-reading. It won't tell you "the truth." It will tell you the possibilities you didn't think of. That distinction matters. There's a whole section on it below.

---

🎬 Demo

🕯️ /mindreader seance — "What did they actually mean?"

You dig up a conversation from three months ago. You still can't figure out what they meant by "whatever, it's fine." Feed it in:

You          ❯ /mindreader seance

mindreader   ❯ Paste the conversation. Then tell me what you're stuck on.

You          ❯ [pastes 20 messages]
             ❯ I can't tell if "whatever, it's fine" was passive-aggressive
             ❯ or if she really didn't care anymore.

mindreader   ❯ I see 3 contradictory readings in these 20 messages.
             ❯ I don't know which one is true. You probably know better than I do.
             ❯
             ❯ ━━━ Reading A: She was angry (confidence ~40%)
             ❯ Evidence: msg #7 she opened up the weekend plan, you replied
             ❯           short. msg #12 ends with a period — in your prior
             ❯           chats, she only uses periods 3 times, all when upset.
             ❯ If this reading is right: she was waiting for you to push.
             ❯
             ❯ ━━━ Reading B: She was actually exhausted (confidence ~35%)
             ❯ Evidence: msg #3 mentions overtime. Reply intervals stretched
             ❯           from ~2min to ~18min by msg #15.
             ❯ If this reading is right: pushing would have made it worse.
             ❯
             ❯ ━━━ Reading C: She wasn't thinking that hard (confidence ~25%)
             ❯ Evidence: 4 prior instances of similar phrasing with no
             ❯           follow-up conflict. You may be projecting current
             ❯           anxiety backward onto a neutral moment.
             ❯
             ❯ ⚠️  I can make all three of these sound true. That's what
             ❯     LLMs are best at, and that's the trap. Treat this as a
             ❯     list of "possibilities I didn't consider," not an answer.

Key design: the shadow won't give you one answer. It gives you three contradictory ones, with evidence, confidence levels, and a self-warning. This is what separates this tool from a Magic 8-Ball. There's a whole section below on why this matters.

---

🎭 /mindreader rehearse — "Run the conversation 100 times before it happens once."

You want to ask your boss for a raise. You also know your boss is the kind of person who shuts down the moment they feel cornered. Run it past the shadow first:

You          ❯ /mindreader rehearse

mindreader   ❯ What's your goal? (one sentence)

You          ❯ Get a 20% raise without damaging the relationship.

mindreader   ❯ How are you planning to open?

You          ❯ "Hey, I think I've contributed a lot this year and I'd
             ❯  like to talk about my comp."

mindreader   ❯ Ran 5 simulated rounds. Goal achievement: 2/10.
             ❯
             ❯ ━━━ Shadow's response
             ❯ Shadow: "Contributed a lot? Like what specifically?"
             ❯ Shadow's inner monologue: (here's another one who thinks
             ❯                            they're indispensable)
             ❯
             ❯ ━━━ Failure mode
             ❯ You triggered their "being extracted from" defense. In their
             ❯ mental model, "I contributed a lot" is a conclusion, not an
             ❯ argument. They want the argument first.
             ❯
             ❯ ━━━ Suggested rewrite (estimated achievement: 7/10)
             ❯ "Hey, I'd love to walk through the projects I shipped this
             ❯  year and talk about where I should focus next year."
             ❯
             ❯ This version routes around the trigger word "raise," reframes
             ❯ the conversation from "I want" to "let's align," which fits
             ❯ their communication style. Bring up comp in round two, after
             ❯ they've already nodded along to the framing.
             ❯
             ❯ ⚠️  The shadow is built from the chat logs you gave me. If
             ❯     the logs are thin, or only cover one type of situation,
             ❯     the simulation will be off. This is a rehearsal, not a
             ❯     script. Stay present in the real conversation.

Key design: every round labels the shadow's "inner monologue," and every output ships with a fidelity warning. Those two details are what separate entertainment from a serious tool.

---

📐 /mindreader calibrate — "How well does my shadow actually know them?"

You've built a shadow of your partner. But how accurate is it, really? Feed in a real conversation and find out:

You          ❯ /mindreader calibrate

mindreader   ❯ Calibration works in two steps. First, paste the
             ❯ FIRST HALF of a real conversation (at least 5 msgs).
             ❯
             ❯ ⚠️  Do NOT paste the full conversation at once.
             ❯     If I see the ending, I'm "remembering," not
             ❯     "predicting." The whole test becomes fake.

You          ❯ [pastes first 15 messages]

mindreader   ❯ Got 15 messages as context. Ready to predict.
             ❯ Say "go" and I'll generate what the shadow thinks
             ❯ happens next.

You          ❯ Go.

mindreader   ❯ ━━━ Shadow's Prediction ━━━
             ❯
             ❯ Based on these 15 messages + shadow profile:
             ❯
             ❯   #1 [them]:  "haha sure sure"
             ❯   #2 [you*]:  "so are we going or not"
             ❯   #3 [them]:  "idk you decide"
             ❯   #4 [you*]:  "fine let's just stay in"
             ❯   #5 [them]:  "k"
             ❯
             ❯   (* = shadow's guess of what you'd say)
             ❯
             ❯ Now paste the REAL second half. Let's see how
             ❯ close it got.

You          ❯ [pastes real messages 16–30]

mindreader   ❯ ━━━ Shadow Calibration Report ━━━
             ❯
             ❯ 📊 Score: 7.2/10
             ❯
             ❯ ━━━ Side-by-side ━━━
             ❯
             ❯ Prediction #1 vs Real #16 (them):
             ❯   Real:   "lol ok whatever you say"
             ❯   Shadow: "haha sure sure"
             ❯   Score: 8/10
             ❯   Analysis: Same dismissive-affectionate tone.
             ❯             Shadow nailed the vibe, missed the
             ❯             specific phrasing.
             ❯
             ❯ Prediction #3 vs Real #22