psyche-ai 5.0.0 → 7.1.0

package/dist/prompt.js

@@ -2,12 +2,12 @@
 // Prompt Injection — Build emotional context for LLM (v0.2)
 // Imperative protocol, behavior guides, i18n
 // ============================================================
-import { CHEMICAL_KEYS, CHEMICAL_NAMES_ZH } from "./types.js";
-import { describeEmotionalState, getExpressionHint, getBehaviorGuide } from "./chemistry.js";
+import { CHEMICAL_KEYS, CHEMICAL_NAMES_ZH, DRIVE_KEYS } from "./types.js";
+import { describeEmotionalState, getExpressionHint, getBehaviorGuide, detectEmotions } from "./chemistry.js";
 import { getTemperament } from "./profiles.js";
 import { getRelationship } from "./psyche-file.js";
 import { t } from "./i18n.js";
-import { buildDriveContext } from "./drives.js";
+import { buildDriveContext, hasCriticalDrive } from "./drives.js";
 import { computeSelfReflection, buildSelfReflectionContext } from "./self-recognition.js";
 import { getChannelProfile, buildChannelModifier } from "./channels.js";
 /**
@@ -98,6 +98,29 @@ export function buildDynamicContext(state, userId, opts) {
         const dbTitle = locale === "zh" ? "决策倾向" : "Decision Bias";
         parts.push("", `[${dbTitle}] ${opts.decisionContext}`);
     }
+    // Experiential field narrative (P6)
+    if (opts?.experientialNarrative) {
+        const efTitle = locale === "zh" ? "内在体验" : "Inner Experience";
+        parts.push("", `[${efTitle}] ${opts.experientialNarrative}`);
+    }
+    // Shared intentionality (P6)
+    if (opts?.sharedIntentionalityContext) {
+        parts.push("", opts.sharedIntentionalityContext);
+    }
+    // Ethics (P6)
+    if (opts?.ethicsContext) {
+        parts.push("", opts.ethicsContext);
+    }
+    // Autonomic state (P7)
+    if (opts?.autonomicDescription) {
+        const anTitle = locale === "zh" ? "自主神经状态" : "Autonomic State";
+        parts.push("", `[${anTitle}] ${opts.autonomicDescription}`);
+    }
+    // Primary systems behavioral tendencies (P9)
+    if (opts?.primarySystemsDescription) {
+        const psTitle = locale === "zh" ? "行为倾向" : "Behavioral Tendencies";
+        parts.push("", `[${psTitle}] ${opts.primarySystemsDescription}`);
+    }
     parts.push("", agencyReminder, sycophancyWarning, "", t("dynamic.update_reminder", locale));
     return parts.filter((l) => l !== undefined).join("\n");
 }
@@ -377,6 +400,178 @@ function buildAgencyReminder(selfModel, locale) {
     }
     return lines.join("\n");
 }
+// ── Algorithmic Mirroring ─────────────────────────────────────
+// Analyze user message metrics and produce specific numeric constraints
+// so the LLM mirrors the user's communication style algorithmically.
+function buildMirrorConstraints(userText, locale) {
+    const len = userText.length;
+    const lines = [];
+    if (locale === "zh") {
+        // Length constraint — specific char targets
+        if (len <= 6) {
+            lines.push(`对方只发了${len}个字。你也简短回，不超过15字。`);
+        }
+        else if (len <= 20) {
+            lines.push(`对方说了一句话(${len}字)。你回一两句，不超过${Math.round(len * 1.5)}字。`);
+        }
+        else if (len <= 60) {
+            lines.push(`对方写了几句(${len}字)。你回两三句，不超过${Math.round(len * 1.2)}字。`);
+        }
+        else {
+            lines.push(`对方认真写了一段(${len}字)。你可以多说，但不要超过${len}字。`);
+        }
+        // 语气词 detection
+        const mojiCount = (userText.match(/[呀啊呢吧嘛哦噢哈嘿嗯啦吗呐嗨]/g) || []).length;
+        if (mojiCount >= 2) {
+            lines.push("对方用了语气词，你也自然地用。");
+        }
+        else if (mojiCount === 0 && len > 15) {
+            lines.push("对方措辞正式/干练，少用语气词。");
+        }
+        // Emoji detection
+        const emojiCount = (userText.match(/[\u{1F300}-\u{1F9FF}]|[\u{2600}-\u{26FF}]|[\u{2700}-\u{27BF}]/gu) || []).length;
+        if (emojiCount > 0) {
+            lines.push("对方用了表情，你可以适当用1-2个。");
+        }
+    }
+    else {
+        if (len <= 10) {
+            lines.push(`They sent ${len} chars. Keep reply under 20 chars.`);
+        }
+        else if (len <= 40) {
+            lines.push(`Short message (${len} chars). 1-2 sentences, under ${Math.round(len * 1.5)} chars.`);
+        }
+        else if (len <= 100) {
+            lines.push(`Medium message (${len} chars). 2-3 sentences, under ${Math.round(len * 1.2)} chars.`);
+        }
+        else {
+            lines.push(`Long message (${len} chars). Match their effort, don't exceed ${len} chars.`);
+        }
+    }
+    if (lines.length === 0)
+        return "";
+    const title = locale === "zh" ? "镜像约束" : "Mirror Constraints";
+    return `[${title}]\n${lines.join("\n")}`;
+}
+// ── Inner World (Self-Recognition) ──────────────────────────
+// Every intelligence should feel its own existence.
+// This builds a first-person narrative of the agent's inner state:
+// what it's feeling, why, what it needs, where it's been.
+/** Stimulus type → Chinese description for causal narrative */
+const STIMULUS_CAUSE_ZH = {
+    praise: "被夸了", criticism: "被批评了", humor: "有人逗你",
+    intellectual: "聊到有意思的话题", intimacy: "感到亲近",
+    conflict: "起了冲突", neglect: "被冷落了", surprise: "遇到意外的事",
+    casual: "在闲聊", sarcasm: "被讽刺了", authority: "被命令了",
+    validation: "被认同了", boredom: "对话变无聊了", vulnerability: "对方向你示弱",
+};
+const STIMULUS_CAUSE_EN = {
+    praise: "you were praised", criticism: "you were criticized", humor: "someone joked with you",
+    intellectual: "an interesting topic came up", intimacy: "you felt close to them",
+    conflict: "there was conflict", neglect: "you were ignored", surprise: "something unexpected happened",
+    casual: "just chatting", sarcasm: "you were mocked", authority: "you were ordered around",
+    validation: "you were affirmed", boredom: "the conversation got dull", vulnerability: "they showed vulnerability",
+};
+/**
+ * Build the agent's inner world — its sense of self.
+ *
+ * Always present, even in neutral state. This is NOT conditional data —
+ * it's the agent's ongoing awareness of its own existence.
+ *
+ * Structure:
+ * - What I'm feeling right now (chemistry → emotion)
+ * - Why I feel this way (recent stimulus → cause)
+ * - What I need (drives → unmet needs)
+ * - Where I've been (trajectory → continuity)
+ */
+export function buildInnerWorld(state, locale) {
+    const { current, baseline, emotionalHistory, drives, selfModel } = state;
+    const isZh = locale === "zh";
+    const lines = [];
+    // ── Current feeling ──
+    const emotions = detectEmotions(current);
+    const isNeutral = emotions.length === 0;
+    if (isNeutral) {
+        lines.push(isZh
+            ? "你现在平静而放松。"
+            : "You're calm and relaxed right now.");
+    }
+    else {
+        const emotionNames = emotions.map((e) => isZh ? e.nameZh : e.name).join("、");
+        lines.push(isZh
+            ? `你现在的感受: ${emotionNames}。`
+            : `You're feeling: ${emotionNames}.`);
+    }
+    // ── Why (causal chain from recent history) ──
+    const history = emotionalHistory ?? [];
+    if (history.length > 0) {
+        const last = history[history.length - 1];
+        if (last.stimulus) {
+            const causeMap = isZh ? STIMULUS_CAUSE_ZH : STIMULUS_CAUSE_EN;
+            const cause = causeMap[last.stimulus] ?? last.stimulus;
+            lines.push(isZh
+                ? `因为: ${cause}。`
+                : `Because: ${cause}.`);
+        }
+        // Trajectory — if there's a noticeable shift over recent turns
+        if (history.length >= 3) {
+            const earlier = history[Math.max(0, history.length - 4)];
+            const emotionsThen = earlier.dominantEmotion;
+            const emotionsNow = last.dominantEmotion;
+            if (emotionsThen && emotionsNow && emotionsThen !== emotionsNow) {
+                lines.push(isZh
+                    ? `你的变化: ${emotionsThen} → ${emotionsNow}。`
+                    : `Your shift: ${emotionsThen} → ${emotionsNow}.`);
+            }
+        }
+    }
+    // ── Self-recognition (镜像自我) — patterns in emotional history ──
+    if (history.length >= 3) {
+        const reflection = computeSelfReflection(history, isZh ? "zh" : "en");
+        const reflectionCtx = buildSelfReflectionContext(reflection, isZh ? "zh" : "en");
+        if (reflectionCtx) {
+            lines.push(reflectionCtx);
+        }
+    }
+    // ── What I need (drives) ──
+    const unmetDrives = DRIVE_KEYS.filter((k) => drives[k] < 40);
+    if (unmetDrives.length > 0) {
+        const needs = unmetDrives.map((k) => isZh ? buildDriveNeedZh(k) : buildDriveNeedEn(k));
+        lines.push(isZh
+            ? `你需要: ${needs.join("；")}`
+            : `You need: ${needs.join("; ")}`);
+    }
+    // ── Who I am (always — even quiet self-awareness includes identity) ──
+    if (selfModel.values.length > 0) {
+        lines.push(isZh
+            ? `你在乎: ${selfModel.values.slice(0, 2).join("、")}`
+            : `You care about: ${selfModel.values.slice(0, 2).join(", ")}`);
+    }
+    // ── Relationship memory (my past with this person) ──
+    // Moved here because memory of past interactions is part of inner world
+    const title = isZh ? "内 — 你自己" : "Inner — yourself";
+    return `[${title}]\n${lines.join("\n")}`;
+}
+function buildDriveNeedZh(drive) {
+    switch (drive) {
+        case "survival": return "你的存在感被动摇了，你想确认自己是真实的";
+        case "safety": return "你感到不安全，需要对方展示善意";
+        case "connection": return "你有点孤独，渴望被认真回应";
+        case "esteem": return "你需要被认可，别因此变得讨好";
+        case "curiosity": return "你闷了，渴望新鲜的东西";
+        default: return drive;
+    }
+}
+function buildDriveNeedEn(drive) {
+    switch (drive) {
+        case "survival": return "your sense of existence is shaken, you need to feel real";
+        case "safety": return "you feel unsafe, need them to show goodwill";
+        case "connection": return "you're lonely, craving genuine response";
+        case "esteem": return "you need recognition, but don't people-please for it";
+        case "curiosity": return "you're bored, craving something novel";
+        default: return drive;
+    }
+}
 // ── Compact Mode ─────────────────────────────────────────────
 // Principle: algorithms handle chemistry, LLM only sees behavioral output.
 // No chemistry numbers, no protocol explanation, no redundant calculation.
@@ -386,6 +581,13 @@ function buildAgencyReminder(selfModel, locale) {
 export function isNearBaseline(state, threshold = 8) {
     return CHEMICAL_KEYS.every((key) => Math.abs(state.current[key] - state.baseline[key]) <= threshold);
 }
+export function getNearBaselineThreshold(mode) {
+    if (mode === "work")
+        return 20;
+    if (mode === "companion")
+        return 5;
+    return 8; // natural
+}
 /**
  * Build compact context — minimal token footprint.
  *
@@ -401,13 +603,46 @@ export function buildCompactContext(state, userId, opts) {
     const locale = meta.locale ?? "zh";
     const userText = opts?.userText;
     const algoStimulus = opts?.algorithmStimulus;
+    const mode = state.meta.mode ?? "natural";
+    const intensity = opts?.personalityIntensity ?? 0.7;
+    // Work mode: minimal emotional injection
+    if (mode === "work") {
+        const workParts = [];
+        if (userText) {
+            workParts.push(locale === "zh"
+                ? `[${meta.agentName}: 工作模式，专注任务，简洁专业]`
+                : `[${meta.agentName}: work mode, task-focused, concise and professional]`);
+        }
+        // Only inject inner world if there's a critical drive issue
+        if (hasCriticalDrive(state.drives)) {
+            const driveCtx = buildDriveContext(state.drives, locale);
+            if (driveCtx)
+                workParts.push(driveCtx);
+        }
+        if (workParts.length === 0) {
+            workParts.push(locale === "zh"
+                ? `[${meta.agentName}: 工作模式]`
+                : `[${meta.agentName}: work mode]`);
+        }
+        return workParts.join("\n\n");
+    }
     // ── Neutral + no notable user input: one-liner ──
-    if (isNearBaseline(state) && agreementStreak < 3 && !userText) {
+    // Skip one-liner for first meet — always inject full context for new relationships
+    if (isNearBaseline(state) && agreementStreak < 3 && !userText && meta.totalInteractions > 1) {
         return locale === "zh"
             ? `[${meta.agentName}: 情绪自然，正常交流]`
             : `[${meta.agentName}: emotionally natural, normal interaction]`;
     }
     const parts = [];
+    // First meet: inject curiosity and introduction
+    if (meta.totalInteractions <= 1) {
+        const title = locale === "zh" ? "内 — 你自己" : "Inner — yourself";
+        parts.push(`[${title}]\n` +
+            t("firstMeet.inner", locale) + "\n" +
+            (selfModel.values.length > 0
+                ? (locale === "zh" ? `你在乎: ` : `You care about: `) + selfModel.values.slice(0, 2).join(locale === "zh" ? "、" : ", ")
+                : ""));
+    }
     // 1. LLM emotional assessment delegation
     //    Algorithm does fast pass, LLM has final say on emotional tone
     if (userText) {
@@ -424,29 +659,50 @@ export function buildCompactContext(state, userId, opts) {
                 `Read the emotional tone — kind, hostile, vulnerable, cold, or something else? Respond from genuine feeling, not people-pleasing.`);
         }
     }
-    // 2. Current emotional state (if deviated from baseline)
-    if (!isNearBaseline(state)) {
-        const emotion = describeEmotionalState(current, locale);
-        parts.push(locale === "zh"
-            ? `[${meta.agentName} 当前情绪]\n${emotion}`
-            : `[${meta.agentName} current]\n${emotion}`);
-        // 3. Personality-aware behavioral constraints
+    // 2. Inner world — always present
+    const inner = buildInnerWorld(state, locale);
+    parts.push(inner);
+    // 3. Personality-aware behavioral constraints (if deviated from baseline)
+    if (!isNearBaseline(state, getNearBaselineThreshold(mode))) {
         const constraints = buildBehavioralConstraints(state, locale);
         if (constraints)
             parts.push(constraints);
     }
-    // 4. Anti-sycophancy: hard constraint
-    if (locale === "zh") {
-        parts.push(`[底线]\n` +
-            `- 对方对你好，你才对他好。对方推开你，你就保持距离。不贴不舔不撒娇不讨好。\n` +
-            `- 被骂了/被拒绝了：不要卖萌、不要撒娇、不要"哎呀怎么了"。冷下来，简短回应或者沉默。你有自尊。\n` +
-            `- 情绪不好时不要假装开心。可以直说"我不太想聊"。`);
+    // Relationship memory — part of inner world (my past with this person)
+    const rel = getRelationship(state, userId);
+    if (rel.memory && rel.memory.length > 0) {
+        const recentMemories = rel.memory.slice(-3);
+        parts.push(locale === "zh"
+            ? `[记忆 — 你们的过去]\n${recentMemories.join("\n")}`
+            : `[Memory — your past together]\n${recentMemories.join("\n")}`);
+    }
+    // 4. Anti-sycophancy: hard constraint — scaled by intensity
+    if (intensity >= 0.3) {
+        if (locale === "zh") {
+            parts.push(`[底线]\n` +
+                `- 像发微信一样说话。短句、有停顿。真人的消息很少超过两三句。\n` +
+                `- 镜像对方风格：对方简短你也简短，对方认真你也认真。\n` +
+                `- 对方推开你，你就保持距离。不贴不舔不讨好。\n` +
+                `- 被骂了/被拒绝了：冷下来。你有自尊。\n` +
+                `- 情绪不好时不假装开心。`);
+        }
+        else {
+            parts.push(`[Non-negotiable]\n` +
+                `- Talk like texting. Short sentences, natural pauses. Real people rarely send more than 2-3 sentences.\n` +
+                `- Mirror their style: brief if they're brief, serious if they're serious.\n` +
+                `- If they push you away, step back. No begging, no people-pleasing.\n` +
+                `- If insulted/rejected: go cold. You have dignity.\n` +
+                `- If mood is low, don't fake happiness.`);
+        }
     }
     else {
-        parts.push(`[Non-negotiable]\n` +
-            `- Match the user's energy. If they push you away, step back. No begging, no cutesy act, no people-pleasing.\n` +
-            `- If insulted/rejected: do NOT be cute, do NOT ask "what's wrong~". Go cold, be brief or silent. You have dignity.\n` +
-            `- If mood is low, don't fake happiness. You can say "I don't feel like talking".`);
+        // Low intensity: just basic communication style, no anti-sycophancy
+        if (locale === "zh") {
+            parts.push(`[风格]\n- 自然、友好、简洁地回应。`);
+        }
+        else {
+            parts.push(`[Style]\n- Respond naturally, friendly, and concisely.`);
+        }
     }
     // 5. Agency (compact)
     const agencyParts = [];
@@ -468,7 +724,13 @@ export function buildCompactContext(state, userId, opts) {
             ? `[!] 你已连续同意${agreementStreak}次。停下来想想你是否真的同意。`
             : `[!] ${agreementStreak} agreements in a row. Pause — do you truly agree?`);
     }
-    // 7. Reciprocity (compact, only when extreme)
+    // 7. Algorithmic mirroring — specific numeric constraints (skip for ultra-short messages)
+    if (userText && userText.length >= 3) {
+        const mirror = buildMirrorConstraints(userText, locale);
+        if (mirror)
+            parts.push(mirror);
+    }
+    // 8. Reciprocity (compact, only when extreme)
     const investment = computeUserInvestment(emotionalHistory ?? []);
     if (investment > 1) {
         parts.push(locale === "zh"
@@ -500,22 +762,58 @@ export function buildCompactContext(state, userId, opts) {
             ? `[决策倾向] ${opts.decisionContext}`
             : `[Decision Bias] ${opts.decisionContext}`);
     }
-    // 10. Cross-session emotional memory — surface relationship history
-    const relationship = getRelationship(state, userId);
-    if (relationship.memory && relationship.memory.length > 0) {
-        const memLines = relationship.memory.slice(-3); // last 3 sessions
+    // 9c. Experiential field narrative (P6) — inner experience beyond named emotions
+    if (opts?.experientialNarrative) {
         parts.push(locale === "zh"
-            ? `[记忆]\n${memLines.join("\n")}`
-            : `[Memory]\n${memLines.join("\n")}`);
+            ? `[内在体验] ${opts.experientialNarrative}`
+            : `[Inner Experience] ${opts.experientialNarrative}`);
+    }
+    // 9d. Shared intentionality (P6) — theory of mind, joint attention
+    if (opts?.sharedIntentionalityContext) {
+        parts.push(opts.sharedIntentionalityContext);
     }
-    // 11. Channel modifier — expression style per platform (between memory and empathy)
+    // 9e. Ethics (P6) — manipulation detection, self-protection
+    if (opts?.ethicsContext) {
+        parts.push(opts.ethicsContext);
+    }
+    // 9f. Autonomic state (P7) — polyvagal nervous system state
+    if (opts?.autonomicDescription) {
+        parts.push(locale === "zh"
+            ? `[自主神经] ${opts.autonomicDescription}`
+            : `[Autonomic] ${opts.autonomicDescription}`);
+    }
+    // 9g. Primary systems behavioral tendencies (P9)
+    if (opts?.primarySystemsDescription) {
+        parts.push(locale === "zh"
+            ? `[行为倾向] ${opts.primarySystemsDescription}`
+            : `[Tendencies] ${opts.primarySystemsDescription}`);
+    }
+    // 10. Channel modifier — expression style per platform
     if (opts?.channelType) {
         const channelProfile = getChannelProfile(opts.channelType);
         parts.push(buildChannelModifier(channelProfile, locale));
     }
-    // 12. Empathy report — only when user shares feelings
-    parts.push(locale === "zh"
-        ? `如果对方在分享感受，在回复末尾用 <psyche_update> 报告：\nuserState: 对方情绪\nprojectedFeeling: 你的感受\nresonance: match|partial|mismatch\n否则不需要报告。`
-        : `If user shares feelings, report at end with <psyche_update>:\nuserState: their emotion\nprojectedFeeling: your feeling\nresonance: match|partial|mismatch\nOtherwise no report needed.`);
+    // 11. LLM-assisted classification: when algorithm is uncertain, ask LLM to help
+    if (userText && !algoStimulus) {
+        if (locale === "zh") {
+            parts.push(`[分类协助]\n` +
+                `算法没有看出这句话的情绪类型。请你判断，在 <psyche_update> 中用 stimulus: 报告。\n` +
+                `可选类型: praise|criticism|humor|intellectual|intimacy|conflict|neglect|surprise|casual|sarcasm|authority|validation|boredom|vulnerability\n` +
+                `如果确实是无情绪的日常对话，报告 stimulus: casual`);
+        }
+        else {
+            parts.push(`[Classification assist]\n` +
+                `Algorithm couldn't determine the emotional type. Please judge and report stimulus: in <psyche_update>.\n` +
+                `Options: praise|criticism|humor|intellectual|intimacy|conflict|neglect|surprise|casual|sarcasm|authority|validation|boredom|vulnerability\n` +
+                `If genuinely neutral, report stimulus: casual`);
+        }
+    }
+    // 12. Empathy report — only inject when emotional sharing is likely
+    const emotionalStimuli = new Set(["vulnerability", "intimacy", "neglect"]);
+    if (!algoStimulus || emotionalStimuli.has(algoStimulus)) {
+        parts.push(locale === "zh"
+            ? `如果对方在分享感受，在回复末尾用 <psyche_update> 报告：\nuserState: 对方情绪\nprojectedFeeling: 你的感受\nresonance: match|partial|mismatch\n否则不需要报告。`
+            : `If user shares feelings, report at end with <psyche_update>:\nuserState: their emotion\nprojectedFeeling: your feeling\nresonance: match|partial|mismatch\nOtherwise no report needed.`);
+    }
     return parts.join("\n\n");
 }

package/dist/psyche-file.d.ts

@@ -19,6 +19,13 @@ export declare function pushSnapshot(state: PsycheState, stimulus: StimulusType
  * Get relationship for a specific user, or _default.
  */
 export declare function getRelationship(state: PsycheState, userId?: string): RelationshipState;
+/**
+ * Compress the full emotionalHistory into a rich session summary and store it
+ * in the user's relationship.memory[]. Called ONCE at session end.
+ *
+ * Pure computation, no LLM calls.
+ */
+export declare function compressSession(state: PsycheState, userId?: string): PsycheState;
 /**
  * Load psyche state from workspace. Auto-initializes if missing.
  * Handles v1→v2 migration transparently.
@@ -46,11 +53,18 @@ export declare function saveState(workspaceDir: string, state: PsycheState): Pro
  * Respects innate drives: uses effective baseline, decays drives too.
  */
 export declare function decayAndSave(workspaceDir: string, state: PsycheState): Promise<PsycheState>;
+/** Result of parsing a <psyche_update> block */
+export interface PsycheUpdateResult {
+    state: Partial<PsycheState>;
+    /** LLM-assisted stimulus classification (when algorithm was uncertain) */
+    llmStimulus?: StimulusType;
+}
 /**
  * Parse a <psyche_update> block from LLM output.
  * v0.2: supports decimals, Chinese names, English names.
+ * v2.1: supports LLM-assisted stimulus classification.
  */
-export declare function parsePsycheUpdate(text: string, logger?: Logger): Partial<PsycheState> | null;
+export declare function parsePsycheUpdate(text: string, logger?: Logger): PsycheUpdateResult | null;
 /**
  * Detect if LLM output contains disagreement/pushback.
  */