behavior-os 0.1.0__tar.gz

behavior_os-0.1.0/.gitignore

@@ -0,0 +1,60 @@
+# Python
+__pycache__/
+*.py[cod]
+*.pyo
+*.pyd
+*.so
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# Distribution / packaging
+dist/
+build/
+*.egg-info/
+*.egg
+
+# Testing
+.pytest_cache/
+.coverage
+htmlcov/
+
+# Type checking
+.mypy_cache/
+.ruff_cache/
+
+# IDE
+.idea/
+.vscode/
+*.swp
+*.swo
+
+# macOS
+.DS_Store
+
+# Environment variables
+.env
+.env.*
+
+# LLM provider keys and runtime wrappers (never commit)
+*.key
+cmini-wrapper
+
+# Character pack backups (created by mindset migrate)
+**/.backup/
+
+# Local registry override
+.agentic-mindset/
+
+# Claude Code
+.claude/
+.claude/workspace/
+.worktrees/
+
+# Local harness context — do not commit
+CLAUDE.md
+
+# aurorie-teams runtime workspace — local only, do not commit
+.claude/workspace/

behavior_os-0.1.0/CHANGELOG.md

@@ -0,0 +1,45 @@
+# BehaviorOS Changelog
+
+All notable changes to this project will be documented in this file.
+
+Format: [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
+
+---
+
+## [1.1] — 2026-03-25
+
+### Added
+- `DecisionFramework.heuristics`, `default_strategy`, `fallback_strategy`, `commitment_policy`
+- `Drive` class — `drives` upgraded from `list[str]` to `list[Drive]` with `name`, `intensity`, `description`; bare strings auto-normalized on load
+- `Trait.confidence` — optional float 0–1
+- `EmotionalTendencies.baseline_mood`, `emotional_range`, `frustration_trigger`, `recovery_pattern`
+- `ConditionalVariant.conjunction` — `"any" | "all"` (default: `"any"`)
+- `BehaviorSchema.decision_control` — `"controlled" | "reactive" | "impulsive"`
+- `ToneAxes` — `formality`, `warmth`, `intensity`, `humor` axes on `VoiceSchema`
+- `MetaSchema.license`, `MetaSchema.visibility`
+- `Source.evidence_level` — `"primary" | "secondary" | "tertiary"`
+- 6 new `Source.type` values: `biography`, `film`, `novel`, `essay`, `letter`, `speech`
+- Schema version constants module (`agentic_mindset/schema/version.py`)
+- Loader version check: warning for 1.0 packs, error for unknown versions
+- `docs/schema-reference.md` — full field reference with runtime-critical classification
+- `docs/migrations/1.0-to-1.1.md` — migration guide
+- All 13 standard library packs enriched with new fields
+
+### Changed
+- `decision_speed` no longer accepts `"impulsive"` — use `decision_control: impulsive` instead (**BREAKING**)
+
+### Fixed
+- `CharacterPack.load()` now uses `pack.meta.id` (attribute) instead of `pack.meta["id"]` (dict subscript)
+
+---
+
+## [1.0] — 2026-03-22
+
+### Added
+- Initial Character Pack schema: `meta`, `mindset`, `personality`, `behavior`, `voice`, `sources`
+- Standard library: 13 characters (7 historical, 6 fictional)
+- Fusion Engine: weighted blend, dominant, sequential strategies
+- Behavior IR pipeline: `ConflictResolver → BehaviorIR → ClaudeRenderer`
+- CLI: `mindset init`, `validate`, `preview`, `list`, `generate`, `run`
+- `--format inject` (Behavior IR path) and `--format text` (FusionEngine path)
+- `--explain` structured YAML output for both paths

behavior_os-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,70 @@
+# Contributing to BehaviorOS
+
+## Character Pack Contributions
+
+### Scope
+
+The standard library accepts:
+- **Historical figures** (deceased persons with documented public records)
+- **Fictional characters** (from literature, anime, games, mythology, or other creative works)
+
+**Living persons are not accepted** into the standard library.
+
+### Requirements
+
+- Minimum 3 distinct source materials in `sources.yaml` (enforced by validator)
+- All 6 schema files present and passing `mindset validate`
+- Sources must be publicly accessible
+- For **fictional characters**: sources must include the original work (manga, novel, screenplay, game); fan-created secondary sources may supplement but not replace primary sources
+- For **historical figures**: primary sources (translated writings, documented speeches) preferred; biographical analysis may supplement
+
+### Process
+
+1. Fork the repository
+2. Scaffold a new pack: `mindset init <id> --type <historical|fictional> --output characters/`
+3. Fill in all 6 YAML files with accurate, sourced content
+4. Add at least 3 sources to `sources.yaml`
+5. Validate: `mindset validate characters/<id>/` — must pass
+6. Preview: `mindset preview characters/<id>/` — review the Context Block output for quality
+7. Open a Pull Request with a brief description of your sources
+
+### What Makes a Good Character Pack?
+
+- Fields are filled from documented evidence, not speculation
+- `confidence` values are set on mindset core_principles to indicate evidence strength
+- `signature_phrases` use direct quotes from primary sources
+- The Context Block preview produces useful, coherent output
+
+## Development
+
+### Setup
+
+```bash
+git clone https://github.com/mengran-aurorie/behavior-os
+cd behavior-os
+pip install -e ".[dev]"
+```
+
+### Running Tests
+
+```bash
+pytest tests/ -v --cov=agentic_mindset
+```
+
+### Project Structure
+
+```
+agentic_mindset/     # Core Python package
+  schema/            # Pydantic schema models
+  ir/                # BehaviorIR models and conditions
+  resolver/          # ConflictResolver and policies
+  renderer/          # ClaudeRenderer (inject path)
+  pack.py            # CharacterPack loader
+  registry.py        # Registry path resolution
+  context.py         # ContextBlock output
+  fusion.py          # FusionEngine
+  cli.py             # Typer CLI
+characters/          # Standard library packs
+tests/               # Test suite (254 tests)
+docs/                # Documentation
+```

behavior_os-0.1.0/PKG-INFO

@@ -0,0 +1,12 @@
+Metadata-Version: 2.4
+Name: behavior-os
+Version: 0.1.0
+Summary: BehaviorOS — compile behavior, not prompts. A behavioral runtime for AI agents.
+Requires-Python: >=3.11
+Requires-Dist: pydantic>=2.0
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: rich>=13.0
+Requires-Dist: typer>=0.12
+Provides-Extra: dev
+Requires-Dist: pytest-cov>=5.0; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'

behavior_os-0.1.0/README.md

@@ -0,0 +1,268 @@
+# BehaviorOS
+
+**Compile behavior, not prompts.**
+
+<p align="center">
+  <a href="./README.md">English</a> | <a href="./README.zh.md">中文</a>
+</p>
+
+<p align="center">
+  <img src="https://img.shields.io/badge/Python-3.11+-blue.svg" alt="Python 3.11+">
+  <img src="https://img.shields.io/badge/License-MIT-green.svg" alt="MIT License">
+  <img src="https://img.shields.io/badge/Schema-1.1-orange.svg" alt="Schema 1.1">
+</p>
+
+> **Naming note:** Product name is **BehaviorOS**. PyPI / package name is `agentic-mindset`. CLI command is `mindset` (preserved for compatibility). Repository is `behavior-os`.
+
+---
+
+## Most AI Personas Are Fake.
+
+> They collapse when you mix them.
+> They can't explain why they behave a certain way.
+> And worse — they start hallucinating identity.
+
+**BehaviorOS** is different.
+
+---
+
+## This Is Not Persona Prompting. This Is Behavior Compilation.
+
+*Personas describe style. Policies define behavior.*
+
+BehaviorOS compiles character mindsets into **behavioral directives** — not character descriptions. The system resolves conflicts, applies conditional rules, and produces outputs you can actually verify.
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  CharacterPack(s)  →  ConflictResolver  →  BehaviorIR           │
+│                                          →  ClaudeRenderer       │
+│                                          →  AI Agent            │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+| Stage | What it does |
+|---|---|
+| **FusionEngine** | Weighted merge of N characters with deterministic conflict resolution |
+| **ConflictResolver** | Slot-by-slot winner selection; ConditionalSlot application |
+| **BehaviorIR** | Typed intermediate representation — every decision is explicit |
+| **ClaudeRenderer** | Emits behavioral directives, not character descriptions |
+
+---
+
+## Same Question, Different Behavior
+
+**Prompt:** *We are negotiating with a much larger competitor. They have more leverage. What do we do?*
+
+---
+
+### Baseline (no persona)
+
+> "You should consider both collaborative and competitive approaches. Assess your BATNA. Look for mutual gains..."
+
+<span style="color:#888">→ generic, balanced, no distinctive frame</span>
+
+---
+
+### Sun Tzu
+
+> "The negotiation is won before the first offer is made. Your goal is not to reach agreement — it is to shape the terrain so agreement favors you."
+
+<span style="color:#c0392b">→ indirect · positioning · no premature commitment</span>
+
+---
+
+### Steve Jobs
+
+> "Stop thinking about leverage. Ask one question: are they necessary? If yes — what's the minimum you need? Cut everything else."
+
+<span style="color:#2980b9">→ direct · binary · refusal to dilute</span>
+
+---
+
+### Sun Tzu (60%) + Steve Jobs (40%)
+
+> "Position first. Never enter a room where the other side has shaped the frame. Find the asymmetry — what do they need? What is their cost of walking? Then act. Precisely, not aggressively."
+
+<span style="color:#27ae60">→ indirect strategy framing + Jobs-style refusal to dilute</span>
+
+> **This exact phrase does not appear in either solo output.** The blend is emergent — not an average.
+
+---
+
+## The System Explains Its Own Behavior
+
+Every run produces a `--explain` YAML that traces every decision:
+
+```yaml
+communication:
+  primary:
+    value: Indirect, layered; teaches through demonstration
+    source: sun-tzu
+    weight: 0.6
+  has_conflict: true
+  dropped:
+    - value: Direct, opinionated, unvarnished
+      source: steve-jobs
+      weight: 0.4
+      reason: no_conflict          ← Jobs' directness doesn't compete with Tzu's indirect
+  modifiers:
+    - value: Direct and uncompromising when clarity_critical
+      condition: [clarity_critical]
+      conjunction: any
+      source: steve-jobs
+      provenance: pack
+```
+
+The `clarity_critical` modifier is a **ConditionalSlot** — Steve Jobs' directness activates only when the situation is already clear. The blend knows when to apply each layer.
+
+> **"The system explains its own behavior — and the explanation matches reality."**
+
+---
+
+## Three Behaviors That Don't Collapse
+
+| Claim | How BehaviorOS delivers it |
+|---|---|
+| **Persona changes output** | Resolver picks winner per slot; renderer enforces it in directives |
+| **Fusion produces emergent behavior** | `no_conflict` drops traits that don't compete; new combinations appear that neither solo has |
+| **Explain predicts output** | Dropped traits are labeled; the benchmark suite verifies they don't surface |
+
+The benchmark suite (`tests/test_benchmark_assertions.py`) verifies all three — including `no fabricated specifics`: the system will not invent biographical facts to fill a persona frame.
+
+### Benchmark Snapshot
+
+| Metric | Value |
+|---|---|
+| Task classes | 4 (persona change, fusion emergence, explain fidelity, no fabrication) |
+| Assertions | 10 |
+| Current status | All green |
+
+---
+
+## Why "OS"?
+
+BehaviorOS is not an operating system in the traditional sense.
+
+It is a **behavioral runtime layer** that:
+- Compiles policies into deterministic behavioral directives
+- Resolves conflicts between competing character traits
+- Injects verifiable, explainable behavior into AI agents
+
+Think of it as an operating system for how agents decide and act — not how they execute code.
+
+---
+
+## Support Matrix
+
+| Tier | Runtime |
+|---|---|
+| **Supported** | Claude CLI |
+| **Experimental** | MiniMax wrapper (`CMINI_WRAPPER_API_KEY`) |
+| **Planned** | OpenAI API · Ollama · any model accepting system prompts |
+
+---
+
+## Contribute a Persona Pack in 3 Steps
+
+1. `mindset init my-character --type historical` — scaffold the YAML files
+2. Fill `sources.yaml` with **3+ public sources** (this is the quality floor)
+3. `mindset validate ./my-character` — verify before submitting
+
+See [CONTRIBUTING.md](./CONTRIBUTING.md) for full authoring guide.
+
+---
+
+## Quick Start
+
+```bash
+# Clone and install
+git clone https://github.com/mengran-aurorie/behavior-os.git
+cd behavior-os
+pip install .
+
+# Or editable mode for development
+pip install -e .
+
+# Single persona
+mindset run claude --persona sun-tzu -- \
+  "We have incomplete data and significant risk. What should we do?"
+
+# Two-persona blend
+mindset run claude --persona sun-tzu --persona steve-jobs --weights 6,4 -- \
+  "We are negotiating with a much larger competitor."
+
+# See every decision made
+mindset run claude --persona sun-tzu --persona steve-jobs --weights 6,4 --explain -- "..."
+```
+
+> **Requires:** Python 3.11+ · [Claude CLI](https://docs.anthropic.com/en/docs/claude-code)
+
+---
+
+## Architecture
+
+```
+CharacterPack/
+│   ├── meta.yaml          # Identity, schema version, license
+│   ├── mindset.yaml       # Principles, decision framework, mental models
+│   ├── personality.yaml    # Traits, emotional tendencies, ConditionalSlots
+│   ├── behavior.yaml       # Work patterns, decision speed, conflict style
+│   ├── voice.yaml          # Tone, vocabulary, signature phrases
+│   └── sources.yaml        # References (3+ public sources required)
+         │
+         ▼
+  FusionEngine
+  (weighted merge, blend/dominant strategy)
+         │
+         ▼
+  ConflictResolver
+  (slot-by-slot winner, ConditionalSlot triggers)
+         │
+         ▼
+  BehaviorIR
+  (typed: primary, dropped, modifiers per slot)
+         │
+         ▼
+  ClaudeRenderer
+  (behavioral directive block)
+         │
+         ▼
+  Agent Runtime
+  (Claude CLI, API, or any model accepting system prompts)
+```
+
+The inject path is **fully deterministic**: identical inputs → identical IR → identical output. No randomness until the final agent prompt.
+
+---
+
+## Standard Library
+
+### Golden Packs (benchmarked in `tests/test_benchmark_assertions.py`)
+
+| ID | Persona | Behavioral signature |
+|---|---|---|
+| `sun-tzu` | Sun Tzu | Strategic positioning over force |
+| `marcus-aurelius` | Marcus Aurelius | Stoic acceptance; control vs. influence |
+| `steve-jobs` | Steve Jobs | Binary quality judgment; refusal to dilute |
+| `sherlock-holmes` | Sherlock Holmes | Deduction from observed anomaly |
+
+### Extended Library
+
+| ID | Persona | Behavioral signature |
+|---|---|---|
+| `confucius` | Confucius | Relationship-based ethics |
+| `seneca` | Seneca | Stoic action; philosophy as practice |
+
+**Build your own:**
+
+```bash
+mindset init my-character --type historical
+# Edit the YAML files
+mindset validate ./my-character
+```
+
+---
+
+## License
+
+MIT

behavior_os-0.1.0/README.zh.md

@@ -0,0 +1,226 @@
+# BehaviorOS
+
+**Compile behavior, not prompts.**
+
+<p align="center">
+  <a href="./README.md">English</a> | <a href="./README.zh.md">中文</a>
+</p>
+
+<p align="center">
+  <img src="https://img.shields.io/badge/Python-3.11+-blue.svg" alt="Python 3.11+">
+  <img src="https://img.shields.io/badge/License-MIT-green.svg" alt="MIT License">
+  <img src="https://img.shields.io/badge/Schema-1.1-orange.svg" alt="Schema 1.1">
+</p>
+
+---
+
+## 大多数 AI Persona 都是假的。
+
+> 混合多个 persona 时，它们会坍缩成泛泛之谈。
+> 它们无法解释自己为什么这样行为。
+> 更糟的是 — 它们开始虚构身份。
+
+**BehaviorOS** 不一样。
+
+---
+
+## 这不是 Persona Prompting。这是 Behavior Compilation。
+
+*Personas 描述风格。Policies 定义行为。*
+
+BehaviorOS 将角色思维方式编译成**行为指令**，而非角色描述。系统解析冲突，应用条件规则，输出可验证。
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  CharacterPack(s)  →  ConflictResolver  →  BehaviorIR            │
+│                                           →  ClaudeRenderer      │
+│                                           →  AI Agent           │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+| 阶段 | 作用 |
+|---|---|
+| **FusionEngine** | N 个角色的加权合并，确定性冲突解析 |
+| **ConflictResolver** | 按槽位选择获胜者；触发 ConditionalSlot |
+| **BehaviorIR** | 类型化中间表示 — 每个决策都显式可见 |
+| **ClaudeRenderer** | 输出行为指令，而非角色描述 |
+
+---
+
+## 同一问题，不同行为
+
+**Prompt：** *我们正在与一个更强大的竞争对手谈判。他们有更多筹码。我们该怎么办？*
+
+---
+
+### Baseline（无 persona）
+
+> "你应该同时考虑合作和竞争策略。评估你的 BATNA。寻找互利方案..."
+
+<span style="color:#888">→ 泛泛而谈，四平八稳，无鲜明立场</span>
+
+---
+
+### 孙子（Sun Tzu）
+
+> "谈判在第一次报价之前就已经赢了。你的目标不是达成协议 — 而是塑造格局，让协议有利于你。"
+
+<span style="color:#c0392b">→ 间接 · 布局 · 不轻易承诺</span>
+
+---
+
+### 史蒂夫·乔布斯（Steve Jobs）
+
+> "别再想筹码了。问一个问题：他们不可或缺吗？如果是——你最少需要他们提供什么？其他全部砍掉。"
+
+<span style="color:#2980b9">→ 直接 · 二元判断 · 绝不稀释</span>
+
+---
+
+### 孙子（60%）+ 乔布斯（40%）
+
+> "先布局。不要进入一个对方已经设好框架的房间。找到不对称性——他们需要什么？他们退出的代价是什么？然后行动。精准地，而不是激进地。"
+
+<span style="color:#27ae60">→ 孙子式战略框架 + 乔布斯式拒绝稀释</span>
+
+> **这句话不出现在任何单角色输出中。** 这是融合产生的涌现行为——不是两者的平均。
+
+---
+
+## 系统会解释自己的行为
+
+每次运行都会生成 `--explain` YAML，追踪每一个决策：
+
+```yaml
+communication:
+  primary:
+    value: Indirect, layered; teaches through demonstration
+    source: sun-tzu
+    weight: 0.6
+  has_conflict: true
+  dropped:
+    - value: Direct, opinionated, unvarnished
+      source: steve-jobs
+      weight: 0.4
+      reason: no_conflict          ← 乔布斯的直接与孙子的间接并不冲突
+  modifiers:
+    - value: Direct and uncompromising when clarity_critical
+      condition: [clarity_critical]
+      conjunction: any
+      source: steve-jobs
+      provenance: pack
+```
+
+`clarity_critical` 是一个 **ConditionalSlot** — 乔布斯的直接风格仅在局势已经明朗时才会激活。融合结果知道何时应用哪一层。
+
+> **"系统会解释自己的行为——而解释与现实完全一致。"**
+
+---
+
+## 三种不会坍缩的行为
+
+| 主张 | BehaviorOS 如何实现 |
+|---|---|
+| **Persona 改变输出** | 解析器逐槽位选优；渲染器以指令形式强制执行 |
+| **融合产生涌现行为** | `no_conflict` 策略丢弃不冲突的特质；新的组合出现，单角色都不具备 |
+| **Explain 预测输出** | 被丢弃的特质有标签标注；基准测试套件验证其不出现 |
+
+基准测试套件（`tests/test_benchmark_assertions.py`）验证以上全部——包括 `no fabricated specifics`：系统不会虚构传记事实来填充 persona 框架。
+
+---
+
+## 为什么叫 "OS"？
+
+BehaviorOS 不是传统意义上的操作系统。
+
+它是一个**行为运行时层**，具备：
+- 将策略编译成确定性行为指令
+- 解析竞争性角色特质之间的冲突
+- 向 AI Agent 注入可验证、可解释的行为
+
+可以把它理解为：管理 Agent 如何决策和行动的"操作系统"——而不是管理代码执行的操作系统。
+
+---
+
+## 快速开始
+
+```bash
+# 安装
+pip install agentic-mindset
+
+# 单角色查询
+mindset run claude --persona sun-tzu -- \
+  "数据不完整且风险很大，我们应该怎么做？"
+
+# 双角色融合
+mindset run claude --persona sun-tzu --persona steve-jobs --weights 6,4 -- \
+  "我们正在与一个更强大的竞争对手谈判。"
+
+# 查看每个决策细节
+mindset run claude --persona sun-tzu --persona steve-jobs --weights 6,4 --explain -- "..."
+```
+
+> **依赖：** Python 3.11+ · [Claude CLI](https://docs.anthropic.com/en/docs/claude-code)
+
+---
+
+## 架构
+
+```
+CharacterPack/
+│   ├── meta.yaml          # 身份、Schema 版本、许可证
+│   ├── mindset.yaml       # 核心原则、决策框架、心智模型
+│   ├── personality.yaml    # 特质、情绪倾向、ConditionalSlot
+│   ├── behavior.yaml       # 工作模式、决策速度、冲突风格
+│   ├── voice.yaml          # 语气、词汇偏好、标志性短语
+│   └── sources.yaml        # 参考资料（至少 3 个公开来源）
+         │
+         ▼
+  FusionEngine
+  （加权合并，blend/dominant 策略）
+         │
+         ▼
+  ConflictResolver
+  （逐槽位裁决，ConditionalSlot 触发）
+         │
+         ▼
+  BehaviorIR
+  （类型化：每个槽位的 primary、dropped、modifiers）
+         │
+         ▼
+  ClaudeRenderer
+  （行为指令块）
+         │
+         ▼
+  Agent Runtime
+  （Claude CLI、API 或任意接受系统提示词的模型）
+```
+
+Inject 路径**完全确定性**：相同输入 → 相同 IR → 相同输出。在最终 agent 提示词之前，无任何随机性。
+
+---
+
+## 标准人物库
+
+| ID | 人物 | 行为特征 |
+|---|---|---|
+| `sun-tzu` | 孙子 | 以布局谋优势，不以力拼 |
+| `marcus-aurelius` | 马可·奥勒留 | 斯多葛接受；区分可控与不可控 |
+| `steve-jobs` | 史蒂夫·乔布斯 | 二元质量判断；拒绝降低标准 |
+| `sherlock-holmes` | 夏洛克·福尔摩斯 | 从观察异常出发进行推理 |
+| `confucius` | 孔子 | 以关系为基础的伦理学 |
+| `seneca` | 塞涅卡 | 斯多葛行动派；哲学即实践 |
+
+**创建你自己的角色：**
+
+```bash
+mindset init my-character --type historical
+# 编辑 YAML 文件
+mindset validate ./my-character
+```
+
+---
+
+## 许可证
+
+MIT

behavior_os-0.1.0/agentic_mindset/__init__.py

@@ -0,0 +1,5 @@
+from agentic_mindset.registry import CharacterRegistry
+from agentic_mindset.fusion import FusionEngine
+from agentic_mindset.context import ContextBlock
+
+__all__ = ["CharacterRegistry", "FusionEngine", "ContextBlock"]