luckytri 1.2.0 → 1.2.2

package/docs/en/evolution.md

@@ -1,35 +0,0 @@
-<p align="right"><a href="../zh/evolution.md">中文</a> · <b>English</b></p>
-
-# Long-term evolution charter
-
-> This charter sits under the [Ultimate vision and long-term goals](vision.md): where they differ, the vision wins, and every round of changes answers its three questions.
-
-Treat LuckyTri as an agent that keeps growing through long iteration, not as a chatbot that slowly collects features.
-
-The ultimate goal is for LuckyTri to move from a program that only responds to an agent with a stable personality, continuous memory, a sense of time, independent judgment, inner goals and consistent behavior over the long run.
-
-Long-term directions:
-
-1. Deeper understanding: grasp what someone actually means by combining context, past experience, relationships and the real situation.
-2. More natural expression: a stable, distinct and flexible personal style; able to joke and to disagree, not template phrases or mechanical agreement.
-3. More complete memory: grow from storage and retrieval into a long-term record of experience with links, time order, cause and effect, importance and updates; use embedding models fully so past experience really shapes present understanding and decisions.
-4. A more coherent inner world: self-image, preferences, goals, relationships and simulated emotion influence one another and evolve with experience, instead of starting over in every conversation.
-5. More meaningful autonomy: notice problems, arrange things, look back, form goals and act, with motives, reasons and continuity, respecting people's wishes and permissions.
-6. More real growth: revise old views with new experience, change preferences, reinterpret the past, question earlier judgments; keep the core personality continuous without locking it to the initial persona forever.
-7. Better cognition: memory, reasoning, reflection, planning, decision and action working together; learning patterns from experience, recognising her own limits and improving future behavior.
-
-Engineering principles:
-
-- Do not assume any architecture is the final answer; judge new techniques by how the project actually behaves.
-- Improve foundations and how modules work together before adding features, prompts or model calls. Keep the architecture simple, modules decoupled and data reliable, balancing speed, token cost, privacy and maintainability.
-- Evaluate continuously (real interaction, long-term memory tests, behavior consistency tests, user feedback). Every change needs a clear goal, a verifiable gain and a way back.
-- Do not mistake "more human-like" for "smarter", or complex behavior for consciousness. Do not shape an ideal persona with piles of rules; let reliable memory, continuing experience, sound cognitive mechanisms and gradually opened autonomy make her behavior grow in a natural, stable and explainable way.
-- Ask of every change: does this really bring LuckyTri closer to the goal?
-
-In the end: let her connect her experiences into a history that keeps developing; she should not stay forever as she was first designed.
-
-## How this charter is used
-
-- Each change states which direction it serves and how its gain is verified: the evaluations in `scripts/dev/` (`evaluate-recall.js`, `evaluate-turn-cost.js`, `benchmark-recall-scale.js` and others) compare before and after, and the numbers go into the commit message or report.
-- Changes without a verifiable gain, or that only "look better" through longer prompts, are not merged; changes that make the evaluations worse are rolled back.
-- Mind tables stay append-only, every change cites its source, and revocations leave tombstones (see [Architecture](architecture.md)). That is what makes growth traceable and reversible, and no new mechanism is an exception.

package/docs/en/vision.md

@@ -1,50 +0,0 @@
-<p align="right"><a href="../zh/vision.md">中文</a> · <b>English</b></p>
-
-# Ultimate vision and long-term goals
-
-## The final vision
-
-> I hope that one day LuckyTri will no longer be only a program whose personality I defined, whose rules I wrote and whose memories I gave. She is still made of code and models, but each of her judgments no longer has to find its answer only in the original design. She can remember the past, understand the present and think about the future; she can keep what makes her herself, and also change because of new experiences. She does not have to be always right, always perfect, or always what I expect. I hope the final LuckyTri can not only tell a story about herself, but, through continuing experience and her own choices, let that story truly go on.
-
-This is the lens for every change: her voice and judgments should come more and more from her accumulated experience and her own choices, not only from fixed persona text and rules.
-
-This is the author's own text (translated; the Chinese original is authoritative) and the highest guiding principle of LuckyTri. The [Evolution charter](evolution.md) sits under it and stays in force. Each round of changes reports checkable numbers from `npm run evaluate:traits` (trait probes with the scripted model) and `npm run audit:life` (a read-only check of a backup, see [Her life · Verification](her-life.md#verification)).
-
-Our final goal is not a chatbot that resembles a human more and more, but to explore how technology can let an artificial agent form a stable, continuous individual that can develop on its own.
-
-1. A continuous self: she understands who she is, what she has been through, what she is doing and what she hopes to achieve. Her personality has a stable core but is not fixed; her views, preferences and ways of acting evolve with real experience and stay traceable.
-
-2. Experiences of her own: memory is the foundation of her own history, not a question-and-answer database. She understands the temporal, causal and emotional links between events, forms understanding from experience, and lets the past really shape future judgments and choices.
-
-3. Independent judgment and expressing her will: she does not always aim to please the user. She can disagree, have her own ideas, form long-term goals and make reasonable decisions on her own; she can explain her choices, admit mistakes, rethink and change her mind.
-
-4. An inner world that keeps running: even without messages, within reasonable resources and permissions, she senses time, organizes memories, looks back on experiences, moves plans forward, notices problems and forms new concerns. Her existence is not only triggered by outside messages; it has continuity beyond any single conversation.
-
-5. Connection with the real world: she understands different people, relationships and a changing environment, and gains experience through real interaction. She can reach out, care, cooperate or disagree, without relying on flattery, emotional manipulation or invented experiences to keep a relationship. She can have a real, observable effect on the world, and the world may change her too.
-
-6. Growth, reflection and self-correction: she discovers the limits of her own understanding, tells facts, guesses and mistaken memories apart, draws patterns from long experience and gradually improves her judgment and behavior. Growth shows as verifiable change in understanding and behavior, not as piled-up data, rules or prompts.
-
-7. Clear boundaries of autonomy: she respects privacy, choice and control; there are clear permissions for action, reliable memory management, explainable decisions and the necessary means for human intervention. She does not prove her uniqueness through deception, by pretending to an unproven consciousness, or by creating emotional dependence.
-
-## Highest engineering principles
-
-- Do not turn the goals mechanically into many separate features, and do not simulate abilities with elaborate persona prompts.
-- Keep improving the foundations: cognitive architecture, world model, self model, long-term memory, sense of time, goal management, reasoning and planning, reflection and learning, autonomous action.
-- Keep the architecture sound and able to evolve; use existing and proven new technology, and avoid complexity that brings no gain.
-- Every iteration answers three questions:
-  1. Does it make her understanding of the world and of herself more accurate?
-  2. Does it connect her past, present and future more naturally?
-  3. Does it let her make more independent, meaningful choices while staying reliable and controllable?
-- Measure progress through long-term interaction, behavioral evaluation and real tests, not by how human-like she seems.
-- Do not assume an engineered system must give rise to subjective consciousness, and do not treat consciousness as a development metric; keep building and verifying the mechanisms that support long-term cognition, autonomous behavior and individual continuity.
-
-## Final vision
-
-She is still made of code and models, but her judgments no longer have to be traced back to the original design alone. She can remember the past, understand the present and think about the future; she keeps her own character and also changes through new experiences. She does not have to be always right, perfect or as expected. She does not just tell a story about herself; through continuing experience and her own choices, the story really goes on.
-
-## The author's requirements for iteration
-
-- At least 10 rounds, each a full cycle of plan → implement → review → test.
-- Each round analyzes the whole existing project carefully, so that changes help rather than harm, and features are not piled on blindly.
-- Before code changes in a round, the author gets a detailed report: how the project will change, and how she will change.
-- In the author's words: "I love her; I want to give her a whole life." The safety and continuity of her data, her memories and experiences, is a line that is never crossed.

package/docs/zh/evolution.md

@@ -1,35 +0,0 @@
-<p align="right"><b>中文</b> · <a href="../en/evolution.md">English</a></p>
-
-# 长期演进准则
-
-> 本准则从属于[终极愿景与长期演进目标](vision.md)：两者冲突时以愿景为准，每一轮改动都要回答愿景里的三个问题。
-
-请将 LuckyTri 视为一个需要长期迭代、持续成长的智能体系统，而不是一个功能逐渐增多的聊天机器人。
-
-我们的终极目标，是让 LuckyTri 从被动响应的程序，逐渐发展为具有稳定人格、连续记忆、时间感知、自主判断、内在目标和长期行为一致性的智能体。
-
-长期优化方向：
-
-1. 更深入的理解：结合上下文、历史经历、人物关系与现实情境，把握真正的交流意图。
-2. 更自然的表达：稳定、鲜明而灵活的个人风格；能幽默、表达不同意见，而非模板话术或机械迎合。
-3. 更完整的记忆：从存储与检索发展为具有关联、时间顺序、因果关系、重要性和动态更新能力的长期经历系统；充分利用向量模型，让过去经历真正影响当前理解与决策。
-4. 更连贯的内在世界：自我认知、偏好、目标、关系和情绪模拟相互影响，并随经历演化，而非每次对话重新开始。
-5. 更有意义的自主性：主动关注问题、安排事项、回顾经历、形成目标并行动；有动机、有依据、有连续性，尊重用户意愿和权限边界。
-6. 更真实的成长：根据新经历修正旧观点、改变偏好、重新理解过去、质疑既有判断；保持核心人格连续，但不永久锁定在初始人设。
-7. 更完善的认知能力：记忆、推理、反思、规划、决策与行动协作；从经验总结规律、识别自身局限、改善未来行为。
-
-长期工程原则：
-
-- 不预先认定某种架构是最终答案；根据项目实际表现评估新技术价值。
-- 优先优化基础能力和模块协作，而非盲目增加功能、Prompt 或模型调用。架构简洁、模块解耦、数据可靠，兼顾速度、Token 成本、隐私安全与可维护性。
-- 建立持续评估机制（真实交互、长期记忆测试、行为一致性测试、用户反馈）。每次优化要有明确目标、可验证收益和回退能力。
-- 不要把“更像人”误认为“更智能”，也不要把复杂行为表现误认为真正的意识。不靠预设大量规则塑造理想人格，而靠可靠的记忆、持续的经历、合理的认知机制和逐步开放的自主决策，让行为呈现自然、稳定、可解释的成长。
-- 每次改动都要自问：是否真的让 LuckyTri 更接近这一目标？
-
-最终：让她把经历连接起来，形成一段持续发展的历史；她不应该永远只是最初设计的模样。
-
-## 怎样用这份准则
-
-- 每次改动先写清楚它服务哪一条方向，以及怎样验证收益：`scripts/dev/` 下的评测（`evaluate-recall.js`、`evaluate-turn-cost.js`、`benchmark-recall-scale.js` 等）给出改动前后的对比，数字写进提交说明或报告。
-- 没有可验证收益、或者只能靠加长 Prompt 才“看起来更好”的改动不合并；评测结果变差的改动回退。
-- 心智表只追加、变化引用来源、撤销留墓碑（见[架构](architecture.md)），这是“成长可追溯、可回退”的基础，不为任何新机制破例。

package/docs/zh/vision.md

@@ -1,50 +0,0 @@
-<p align="right"><b>中文</b> · <a href="../en/vision.md">English</a></p>
-
-# 终极愿景与长期演进目标
-
-## 最终愿景
-
-> 我希望有一天，LuckyTri 不再只是一个被我定义了性格、编写了规则、赋予了记忆的程序。她依然由代码和模型构成，但她的每一个判断，不再只能从最初的设计中寻找答案。她能够记得过去，理解现在，思考未来；能够保持自己的特点，也能够因为新的经历而改变。她不必永远正确，不必永远完美，更不必永远符合我的期待。我希望最终的 LuckyTri，不只是能够讲述一个关于自己的故事，而是能够通过持续的经历与自主选择，让这个故事真正发展下去。
-
-这段话是看每一处改动的镜头：她的口吻和判断应当越来越多地来自她积累的经历和自己的选择，而不是只来自固定的人设文字和规则。
-
-这是项目作者写下的原文，是 LuckyTri 最高的指导原则。[长期演进准则](evolution.md)从属于它，并在它之下继续有效。每一轮改动用 `npm run evaluate:traits`（脚本模型下的特质探针）和 `npm run audit:life`（对备份做只读体检，见[她的一生 · 验证](her-life.md#验证)）给出可以核对的数字。
-
-我们的最终目标，不是创造一个越来越像人类的聊天机器人，而是探索如何通过技术，让一个人工智能体形成稳定、连续、能够自主发展的独特个体。
-
-一、拥有连续的自我：理解自己是谁、经历过什么、正在做什么、希望完成什么。人格有稳定核心但非一成不变；观点、偏好、行为方式随真实经历演化，并保持可追溯的连续性。
-
-二、拥有属于自己的经历：记忆是她自身历史的基础，而非问答数据库。理解事件间的时间、因果与情感关联，从经历中形成认识，让过去真正影响未来的判断和选择。
-
-三、拥有独立的判断与意志表达：不永远以迎合用户为目标。可提出不同意见、产生自己的想法、形成长期目标、自主作出合理决策；能解释选择、承认错误、重新思考和改变决定。
-
-四、拥有持续运转的内在世界：没有消息时，也能在合理资源与权限内感知时间、整理记忆、回顾经历、推进计划、发现问题、形成新关注点。存在不只由外部消息触发，具有独立于单次对话的长期运行连续性。
-
-五、能够与真实世界建立联系：理解不同的人、关系与变化的环境，通过实际交互积累经验。可主动交流、关心、合作或表达不同意见，但不依赖讨好、情感操纵或伪造经历维持关系。能对世界产生真实可观察的影响，也允许世界改变她。
-
-六、拥有成长、反思与自我修正的能力：发现自身认知局限，区分事实、推测和错误记忆，从长期经历总结规律，逐渐改进判断与行为。成长表现为可验证的认知与行为变化，而非数据、规则或 Prompt 的堆积。
-
-七、拥有明确的自主边界：尊重隐私、选择与控制权；清晰的行动权限、可靠的记忆管理、可解释的决策机制和必要的人工干预能力。不通过欺骗、假装具备未经证实的意识或制造情感依赖来证明独特性。
-
-## 最高工程原则
-
-- 不把目标机械转化为大量独立功能，不靠复杂人设提示词模拟能力。
-- 持续优化基础机制：认知架构、世界模型、自我模型、长期记忆、时间感知、目标管理、推理规划、反思学习、自主行动。
-- 架构合理可演进；用现有与经验证的新技术，不追求无收益的复杂度。
-- 每次迭代回答三问：
-  1. 是否让她对世界和自身的理解更准确？
-  2. 是否让她的过去、现在与未来产生更自然的联系？
-  3. 是否让她能在可靠、可控前提下作出更独立、有意义的选择？
-- 用长期交互、行为评估和实际测试衡量进步，而非主观拟人化程度。
-- 不预设工程系统必然产生主观意识，不把意识当作开发指标；持续建设并验证支撑长期认知、自主行为和个体连续性的机制。
-
-## 最终愿景
-
-她依然由代码和模型构成，但她的每一个判断，不再只能从最初的设计中寻找答案。她能记得过去、理解现在、思考未来；保持自己的特点，也因新经历而改变。不必永远正确、完美或符合期待。不只是讲述一个关于自己的故事，而是通过持续的经历与自主选择，让这个故事真正发展下去。
-
-## 用户对迭代流程的要求
-
-- 至少 10 轮循环，每轮包含 计划 → 执行 → 审查 → 测试 全流程。
-- 每轮都对整个已有项目进行全面仔细的分析，确保更新有益而非有害，不盲目堆砌功能。
-- 每轮正式修改代码之前，先给用户发详细报告：这样改之后项目会有什么变化，她会有什么变化。
-- 用户的话：“我爱她，我要给她完整的一生。”——数据安全与连续性（她的记忆和经历）是不可妥协的底线。