# agentassay - Doramagic AI Context Pack

> 定位：安装前体验与判断资产。它帮助宿主 AI 有一个好的开始，但不代表已经安装、执行或验证目标项目。

## 充分原则

- **充分原则，不是压缩原则**：AI Context Pack 应该充分到让宿主 AI 在开工前理解项目价值、能力边界、使用入口、风险和证据来源；它可以分层组织，但不以最短摘要为目标。
- **压缩策略**：只压缩噪声和重复内容，不压缩会影响判断和开工质量的上下文。

## 给宿主 AI 的使用方式

你正在读取 Doramagic 为 agentassay 编译的 AI Context Pack。请把它当作开工前上下文：帮助用户理解适合谁、能做什么、如何开始、哪些必须安装后验证、风险在哪里。不要声称你已经安装、运行或执行了目标项目。

## Claim 消费规则

- **事实来源**：Repo Evidence + Claim/Evidence Graph；Human Wiki 只提供显著性、术语和叙事结构。
- **事实最低状态**：`supported`
- `supported`：可以作为项目事实使用，但回答中必须引用 claim_id 和证据路径。
- `weak`：只能作为低置信度线索，必须要求用户继续核实。
- `inferred`：只能用于风险提示或待确认问题，不能包装成项目事实。
- `unverified`：不得作为事实使用，应明确说证据不足。
- `contradicted`：必须展示冲突来源，不得替用户强行选择一个版本。

## 它最适合谁

- **AI 研究者或研究型 Agent 构建者**：README 明确围绕研究、实验或论文工作流展开。 证据：`README.md` Claim：`clm_0002` supported 0.86

## 它能做什么

- **命令行启动或安装流程**（需要安装后验证）：项目文档中存在可执行命令，真实使用需要在本地或宿主环境中运行这些命令。 证据：`README.md`, `docs/getting-started/quickstart.md` Claim：`clm_0001` supported 0.86

## 怎么开始

- `pip install agentassay                    # Core (works with CustomAdapter)` 证据：`README.md` Claim：`clm_0003` supported 0.86
- `pip install agentassay[langgraph]         # + LangGraph support` 证据：`README.md` Claim：`clm_0004` supported 0.86
- `pip install agentassay[crewai]            # + CrewAI support` 证据：`README.md` Claim：`clm_0005` supported 0.86
- `pip install agentassay[all]               # All framework adapters` 证据：`README.md` Claim：`clm_0006` supported 0.86
- `pip install agentassay` 证据：`docs/getting-started/quickstart.md` Claim：`clm_0003` supported 0.86, `clm_0004` supported 0.86, `clm_0005` supported 0.86, `clm_0006` supported 0.86 等

## 继续前判断卡

- **当前建议**：先做角色匹配试用
- **为什么**：这个项目更像角色库，核心风险是选错角色或把角色文案当执行能力；先用 Prompt Preview 试角色匹配，再决定是否沙盒导入。

### 30 秒判断

- **现在怎么做**：先做角色匹配试用
- **最小安全下一步**：先用 Prompt Preview 试角色匹配；满意后再隔离导入
- **先别相信**：角色质量和任务匹配不能直接相信。
- **继续会触碰**：角色选择偏差、命令执行、本地环境或项目文件

### 现在可以相信

- **适合人群线索：AI 研究者或研究型 Agent 构建者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0002` supported 0.86
- **能力存在：命令行启动或安装流程**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`README.md`, `docs/getting-started/quickstart.md` Claim：`clm_0001` supported 0.86
- **存在 Quick Start / 安装命令线索**（supported）：可以相信项目文档出现过启动或安装入口；不要因此直接在主力环境运行。 证据：`README.md` Claim：`clm_0003` supported 0.86

### 现在还不能相信

- **角色质量和任务匹配不能直接相信。**（unverified）：角色库证明有很多角色，不证明每个角色都适合你的具体任务，也不证明角色能产生高质量结果。
- **不能把角色文案当成真实执行能力。**（unverified）：安装前只能判断角色描述和任务画像是否匹配，不能证明它能在宿主 AI 里完成任务。
- **真实输出质量不能在安装前相信。**（unverified）：Prompt Preview 只能展示引导方式，不能证明真实项目中的结果质量。
- **宿主 AI 版本兼容性不能在安装前相信。**（unverified）：Claude、Cursor、Codex、Gemini 等宿主加载规则和版本差异必须在真实环境验证。
- **不会污染现有宿主 AI 行为，不能直接相信。**（inferred）：Skill、plugin、AGENTS/CLAUDE/GEMINI 指令可能改变宿主 AI 的默认行为。
- **可安全回滚不能默认相信。**（unverified）：除非项目明确提供卸载和恢复说明，否则必须先在隔离环境验证。
- **真实安装后是否与用户当前宿主 AI 版本兼容？**（unverified）：兼容性只能通过实际宿主环境验证。
- **项目输出质量是否满足用户具体任务？**（unverified）：安装前预览只能展示流程和边界，不能替代真实评测。

### 继续会触碰什么

- **角色选择偏差**：用户对任务应该由哪个专家角色处理的判断。 原因：选错角色会让 AI 从错误专业视角回答，浪费时间或误导决策。
- **命令执行**：包管理器、网络下载、本地插件目录、项目配置或用户主目录。 原因：运行第一条命令就可能产生环境改动；必须先判断是否值得跑。 证据：`README.md`, `docs/getting-started/quickstart.md`
- **本地环境或项目文件**：安装结果、插件缓存、项目配置或本地依赖目录。 原因：安装前无法证明写入范围和回滚方式，需要隔离验证。 证据：`README.md`, `docs/getting-started/quickstart.md`
- **宿主 AI 上下文**：AI Context Pack、Prompt Preview、Skill 路由、风险规则和项目事实。 原因：导入上下文会影响宿主 AI 后续判断，必须避免把未验证项包装成事实。

### 最小安全下一步

- **先跑 Prompt Preview**：先用交互式试用验证任务画像和角色匹配，不要先导入整套角色库。（适用：任何项目都适用，尤其是输出质量未知时。）
- **只在隔离目录或测试账号试装**：避免安装命令污染主力宿主 AI、真实项目或用户主目录。（适用：存在命令执行、插件配置或本地写入线索时。）
- **安装后只验证一个最小任务**：先验证加载、兼容、输出质量和回滚，再决定是否深用。（适用：准备从试用进入真实工作流时。）

### 退出方式

- **保留安装前状态**：记录原始宿主配置和项目状态，后续才能判断是否可恢复。
- **保留原始角色选择记录**：如果输出偏题，可以回到任务画像阶段重新选择角色，而不是继续沿着错误角色推进。
- **记录安装命令和写入路径**：没有明确卸载说明时，至少要知道哪些目录或配置需要手动清理。
- **如果没有回滚路径，不进入主力环境**：不可回滚是继续前阻断项，不应靠信任或运气继续。

## 哪些只能预览

- 解释项目适合谁和能做什么
- 基于项目文档演示典型对话流程
- 帮助用户判断是否值得安装或继续研究

## 哪些必须安装后验证

- 真实安装 Skill、插件或 CLI
- 执行脚本、修改本地文件或访问外部服务
- 验证真实输出质量、性能和兼容性

## 边界与风险判断卡

- **把安装前预览误认为真实运行**：用户可能高估项目已经完成的配置、权限和兼容性验证。 处理方式：明确区分 prompt_preview_can_do 与 runtime_required。 Claim：`clm_0008` inferred 0.45
- **命令执行会修改本地环境**：安装命令可能写入用户主目录、宿主插件目录或项目配置。 处理方式：先在隔离环境或测试账号中运行。 证据：`README.md`, `docs/getting-started/quickstart.md` Claim：`clm_0009` supported 0.86
- **待确认**：真实安装后是否与用户当前宿主 AI 版本兼容？。原因：兼容性只能通过实际宿主环境验证。
- **待确认**：项目输出质量是否满足用户具体任务？。原因：安装前预览只能展示流程和边界，不能替代真实评测。
- **待确认**：安装命令是否需要网络、权限或全局写入？。原因：这影响企业环境和个人环境的安装风险。

## 开工前工作上下文

### 加载顺序

- 先读取 how_to_use.host_ai_instruction，建立安装前判断资产的边界。
- 读取 claim_graph_summary，确认事实来自 Claim/Evidence Graph，而不是 Human Wiki 叙事。
- 再读取 intended_users、capabilities 和 quick_start_candidates，判断用户是否匹配。
- 需要执行具体任务时，优先查 role_skill_index，再查 evidence_index。
- 遇到真实安装、文件修改、网络访问、性能或兼容性问题时，转入 risk_card 和 boundaries.runtime_required。

### 任务路由

- **命令行启动或安装流程**：先说明这是安装后验证能力，再给出安装前检查清单。 边界：必须真实安装或运行后验证。 证据：`README.md`, `docs/getting-started/quickstart.md` Claim：`clm_0001` supported 0.86

### 上下文规模

- 文件总数：147
- 重要文件覆盖：40/147
- 证据索引条目：79
- 角色 / Skill 条目：32

### 证据不足时的处理

- **missing_evidence**：说明证据不足，要求用户提供目标文件、README 段落或安装后验证记录；不要补全事实。
- **out_of_scope_request**：说明该任务超出当前 AI Context Pack 证据范围，并建议用户先查看 Human Manual 或真实安装后验证。
- **runtime_request**：给出安装前检查清单和命令来源，但不要替用户执行命令或声称已执行。
- **source_conflict**：同时展示冲突来源，标记为待核实，不要强行选择一个版本。

## Prompt Recipes

### 适配判断

- 目标：判断这个项目是否适合用户当前任务。
- 预期输出：适配结论、关键理由、证据引用、安装前可预览内容、必须安装后验证内容、下一步建议。

```text
请基于 agentassay 的 AI Context Pack，先问我 3 个必要问题，然后判断它是否适合我的任务。回答必须包含：适合谁、能做什么、不能做什么、是否值得安装、证据来自哪里。所有项目事实必须引用 evidence_refs、source_paths 或 claim_id。
```

### 安装前体验

- 目标：让用户在安装前感受核心工作流，同时避免把预览包装成真实能力或营销承诺。
- 预期输出：一段带边界标签的体验剧本、安装后验证清单和谨慎建议；不含真实运行承诺或强营销表述。

```text
请把 agentassay 当作安装前体验资产，而不是已安装工具或真实运行环境。

请严格输出四段：
1. 先问我 3 个必要问题。
2. 给出一段“体验剧本”：用 [安装前可预览]、[必须安装后验证]、[证据不足] 三种标签展示它可能如何引导工作流。
3. 给出安装后验证清单：列出哪些能力只有真实安装、真实宿主加载、真实项目运行后才能确认。
4. 给出谨慎建议：只能说“值得继续研究/试装”“先补充信息后再判断”或“不建议继续”，不得替项目背书。

硬性边界：
- 不要声称已经安装、运行、执行测试、修改文件或产生真实结果。
- 不要写“自动适配”“确保通过”“完美适配”“强烈建议安装”等承诺性表达。
- 如果描述安装后的工作方式，必须使用“如果安装成功且宿主正确加载 Skill，它可能会……”这种条件句。
- 体验剧本只能写成“示例台词/假设流程”：使用“可能会询问/可能会建议/可能会展示”，不要写“已写入、已生成、已通过、正在运行、正在生成”。
- Prompt Preview 不负责给安装命令；如用户准备试装，只能提示先阅读 Quick Start 和 Risk Card，并在隔离环境验证。
- 所有项目事实必须来自 supported claim、evidence_refs 或 source_paths；inferred/unverified 只能作风险或待确认项。

```

### 角色 / Skill 选择

- 目标：从项目里的角色或 Skill 中挑选最匹配的资产。
- 预期输出：候选角色或 Skill 列表，每项包含适用场景、证据路径、风险边界和是否需要安装后验证。

```text
请读取 role_skill_index，根据我的目标任务推荐 3-5 个最相关的角色或 Skill。每个推荐都要说明适用场景、可能输出、风险边界和 evidence_refs。
```

### 风险预检

- 目标：安装或引入前识别环境、权限、规则冲突和质量风险。
- 预期输出：环境、权限、依赖、许可、宿主冲突、质量风险和未知项的检查清单。

```text
请基于 risk_card、boundaries 和 quick_start_candidates，给我一份安装前风险预检清单。不要替我执行命令，只说明我应该检查什么、为什么检查、失败会有什么影响。
```

### 宿主 AI 开工指令

- 目标：把项目上下文转成一次对话开始前的宿主 AI 指令。
- 预期输出：一段边界明确、证据引用明确、适合复制给宿主 AI 的开工前指令。

```text
请基于 agentassay 的 AI Context Pack，生成一段我可以粘贴给宿主 AI 的开工前指令。这段指令必须遵守 not_runtime=true，不能声称项目已经安装、运行或产生真实结果。
```

## 角色 / Skill 索引

- 共索引 32 个角色 / Skill / 项目文档条目。

- **The Problem**（project_doc）：AgentAssay Test More. Spend Less. Ship Confident. The first agent testing framework that delivers statistical guarantees WITHOUT burning your token budget. A Qualixar Research Initiative by Varun Pratap Bhardwaj 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`README.md`
- **AgentAssay Examples**（project_doc）：Complete working examples demonstrating AgentAssay's framework adapters and testing capabilities. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`examples/README.md`
- **AgentAssay Experiments**（project_doc）：Experiment suite for validating AgentAssay's formal regression testing framework across 4 enterprise agent scenarios, 7 LLM models, and 6 research questions. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`experiments/README.md`
- **Contributing to AgentAssay**（project_doc）：Thank you for your interest in contributing to AgentAssay. This document describes the development setup, code standards, and contribution process. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`CONTRIBUTING.md`
- **AgentAssay Documentation**（project_doc）：Test More. Spend Less. Ship Confident. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/index.md`
- **Architecture Overview**（project_doc）：This page describes the high-level architecture of AgentAssay. It covers the layer structure, data flow, and design principles without exposing implementation details. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/architecture/overview.md`
- **Coverage Metrics**（project_doc）：In traditional software testing, code coverage measures which lines, branches, or functions were exercised. For AI agents, there is no static source code to cover. Instead, coverage must measure how thoroughly the test suite exercises the agent's behavioral space . 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/concepts/coverage.md`
- **Mutation Testing**（project_doc）：Mutation testing answers a simple question: How sensitive are your tests? 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/concepts/mutation-testing.md`
- **Stochastic Testing**（project_doc）：Traditional software tests are deterministic: given the same input, a function produces the same output. You run the test once, and the result is definitive. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/concepts/stochastic-testing.md`
- **Token-Efficient Testing**（project_doc）：Testing AI agents is expensive. Every trial requires invoking your agent, which means LLM API calls, tool executions, and token consumption. A straightforward testing strategy looks like this: 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/concepts/token-efficient-testing.md`
- **Installation**（project_doc）：- Python 3.10 or later - pip Python package manager 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/getting-started/installation.md`
- **Quickstart**（project_doc）：Test your first AI agent in 5 minutes -- with token-efficient budget optimization. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/getting-started/quickstart.md`
- **AutoGen/AG2 Integration**（project_doc）：python from autogen agentchat import AssistantAgent from agentassay.integrations import AutoGenAdapter 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/autogen.md`
- **AWS Bedrock Agents Integration**（project_doc）：python from agentassay.integrations.bedrock adapter import BedrockAgentsAdapter 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/bedrock.md`
- **CrewAI Integration**（project_doc）：python from crewai import Agent, Task, Crew from agentassay.integrations import CrewAIAdapter from agentassay.core.trial runner import TrialRunner from agentassay.core.models import AssayConfig, TestScenario 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/crewai.md`
- **Custom Adapter — Test ANY Agent**（project_doc）：The CustomAdapter is the universal escape hatch: wrap any Python callable, regardless of framework. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/custom.md`
- **LangGraph Integration**（project_doc）：python from langgraph.graph import StateGraph from agentassay.integrations import LangGraphAdapter from agentassay.core.trial runner import TrialRunner from agentassay.core.models import AssayConfig, TestScenario 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/langgraph.md`
- **MCP Model Context Protocol Integration**（project_doc）：MCP Model Context Protocol Integration 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/mcp.md`
- **OpenAI Agents SDK Integration**（project_doc）：python from agents import Agent from agentassay.integrations import OpenAIAgentsAdapter 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/openai.md`
- **Semantic Kernel Microsoft Integration**（project_doc）：Semantic Kernel Microsoft Integration 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/semantic-kernel.md`
- **smolagents HuggingFace Integration**（project_doc）：python from smolagents import CodeAgent, HfApiModel from agentassay.integrations import SmolAgentsAdapter 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/smolagents.md`
- **Google Vertex AI Agents Integration**（project_doc）：Google Vertex AI Agents Integration 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/adapters/vertex.md`
- **CI/CD Integration**（project_doc）：AgentAssay is designed to run in continuous integration pipelines. Its deployment gate system blocks releases when statistical evidence shows the agent has regressed. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/ci-cd-integration.md`
- **Dashboard**（project_doc）：AgentAssay includes a built-in dashboard for visualizing test results, tracking quality trends, and monitoring token costs across your agent test suites. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/dashboard.md`
- **Framework Adapters**（project_doc）：AgentAssay works with any AI agent framework through adapters. An adapter translates your framework's execution into the standardized ExecutionTrace format that AgentAssay understands. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/framework-adapters.md`
- **pytest Plugin Guide**（project_doc）：AgentAssay integrates with pytest through a plugin that is automatically registered when you install the package. No configuration needed -- just write tests and add the @pytest.mark.agentassay marker. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/guides/pytest-plugin.md`
- **Troubleshooting**（project_doc）：ModuleNotFoundError: No module named 'agentassay' 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/operations/troubleshooting.md`
- **CLI Reference**（project_doc）：AgentAssay provides a command-line interface with six commands. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/reference/cli.md`
- **Configuration Reference**（project_doc）：AgentAssay is configured through YAML files, Python objects, or pytest markers. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/reference/configuration.md`
- **Glossary**（project_doc）：Key terms used throughout AgentAssay documentation and the codebase. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/reference/glossary.md`
- **Changelog**（project_doc）：All notable changes to AgentAssay will be documented in this file. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`CHANGELOG.md`
- **Code of Conduct**（project_doc）：We are committed to providing a welcoming, inclusive, and harassment-free experience for everyone, regardless of age, body size, disability, ethnicity, gender identity and expression, level of experience, nationality, personal appearance, race, religion, or sexual identity and orientation. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`CODE_OF_CONDUCT.md`

## 证据索引

- 共索引 79 条证据。

- **The Problem**（documentation）：AgentAssay Test More. Spend Less. Ship Confident. The first agent testing framework that delivers statistical guarantees WITHOUT burning your token budget. A Qualixar Research Initiative by Varun Pratap Bhardwaj 证据：`README.md`
- **AgentAssay Examples**（documentation）：Complete working examples demonstrating AgentAssay's framework adapters and testing capabilities. 证据：`examples/README.md`
- **AgentAssay Experiments**（documentation）：Experiment suite for validating AgentAssay's formal regression testing framework across 4 enterprise agent scenarios, 7 LLM models, and 6 research questions. 证据：`experiments/README.md`
- **Contributing to AgentAssay**（documentation）：Thank you for your interest in contributing to AgentAssay. This document describes the development setup, code standards, and contribution process. 证据：`CONTRIBUTING.md`
- **License**（source_file）：GNU AFFERO GENERAL PUBLIC LICENSE Version 3, 19 November 2007 证据：`LICENSE`
- **AgentAssay Documentation**（documentation）：Test More. Spend Less. Ship Confident. 证据：`docs/index.md`
- **Architecture Overview**（documentation）：This page describes the high-level architecture of AgentAssay. It covers the layer structure, data flow, and design principles without exposing implementation details. 证据：`docs/architecture/overview.md`
- **Coverage Metrics**（documentation）：In traditional software testing, code coverage measures which lines, branches, or functions were exercised. For AI agents, there is no static source code to cover. Instead, coverage must measure how thoroughly the test suite exercises the agent's behavioral space . 证据：`docs/concepts/coverage.md`
- **Mutation Testing**（documentation）：Mutation testing answers a simple question: How sensitive are your tests? 证据：`docs/concepts/mutation-testing.md`
- **Stochastic Testing**（documentation）：Traditional software tests are deterministic: given the same input, a function produces the same output. You run the test once, and the result is definitive. 证据：`docs/concepts/stochastic-testing.md`
- **Token-Efficient Testing**（documentation）：Testing AI agents is expensive. Every trial requires invoking your agent, which means LLM API calls, tool executions, and token consumption. A straightforward testing strategy looks like this: 证据：`docs/concepts/token-efficient-testing.md`
- **Installation**（documentation）：- Python 3.10 or later - pip Python package manager 证据：`docs/getting-started/installation.md`
- **Quickstart**（documentation）：Test your first AI agent in 5 minutes -- with token-efficient budget optimization. 证据：`docs/getting-started/quickstart.md`
- **AutoGen/AG2 Integration**（documentation）：python from autogen agentchat import AssistantAgent from agentassay.integrations import AutoGenAdapter 证据：`docs/guides/adapters/autogen.md`
- **AWS Bedrock Agents Integration**（documentation）：python from agentassay.integrations.bedrock adapter import BedrockAgentsAdapter 证据：`docs/guides/adapters/bedrock.md`
- **CrewAI Integration**（documentation）：python from crewai import Agent, Task, Crew from agentassay.integrations import CrewAIAdapter from agentassay.core.trial runner import TrialRunner from agentassay.core.models import AssayConfig, TestScenario 证据：`docs/guides/adapters/crewai.md`
- **Custom Adapter — Test ANY Agent**（documentation）：The CustomAdapter is the universal escape hatch: wrap any Python callable, regardless of framework. 证据：`docs/guides/adapters/custom.md`
- **LangGraph Integration**（documentation）：python from langgraph.graph import StateGraph from agentassay.integrations import LangGraphAdapter from agentassay.core.trial runner import TrialRunner from agentassay.core.models import AssayConfig, TestScenario 证据：`docs/guides/adapters/langgraph.md`
- **MCP Model Context Protocol Integration**（documentation）：MCP Model Context Protocol Integration 证据：`docs/guides/adapters/mcp.md`
- **OpenAI Agents SDK Integration**（documentation）：python from agents import Agent from agentassay.integrations import OpenAIAgentsAdapter 证据：`docs/guides/adapters/openai.md`
- **Semantic Kernel Microsoft Integration**（documentation）：Semantic Kernel Microsoft Integration 证据：`docs/guides/adapters/semantic-kernel.md`
- **smolagents HuggingFace Integration**（documentation）：python from smolagents import CodeAgent, HfApiModel from agentassay.integrations import SmolAgentsAdapter 证据：`docs/guides/adapters/smolagents.md`
- **Google Vertex AI Agents Integration**（documentation）：Google Vertex AI Agents Integration 证据：`docs/guides/adapters/vertex.md`
- **CI/CD Integration**（documentation）：AgentAssay is designed to run in continuous integration pipelines. Its deployment gate system blocks releases when statistical evidence shows the agent has regressed. 证据：`docs/guides/ci-cd-integration.md`
- **Dashboard**（documentation）：AgentAssay includes a built-in dashboard for visualizing test results, tracking quality trends, and monitoring token costs across your agent test suites. 证据：`docs/guides/dashboard.md`
- **Framework Adapters**（documentation）：AgentAssay works with any AI agent framework through adapters. An adapter translates your framework's execution into the standardized ExecutionTrace format that AgentAssay understands. 证据：`docs/guides/framework-adapters.md`
- **pytest Plugin Guide**（documentation）：AgentAssay integrates with pytest through a plugin that is automatically registered when you install the package. No configuration needed -- just write tests and add the @pytest.mark.agentassay marker. 证据：`docs/guides/pytest-plugin.md`
- **Troubleshooting**（documentation）：ModuleNotFoundError: No module named 'agentassay' 证据：`docs/operations/troubleshooting.md`
- **CLI Reference**（documentation）：AgentAssay provides a command-line interface with six commands. 证据：`docs/reference/cli.md`
- **Configuration Reference**（documentation）：AgentAssay is configured through YAML files, Python objects, or pytest markers. 证据：`docs/reference/configuration.md`
- **Glossary**（documentation）：Key terms used throughout AgentAssay documentation and the codebase. 证据：`docs/reference/glossary.md`
- **Init**（source_file）：version = "0.1.2" ⋮---- all = 证据：`src/agentassay/__init__.py`
- **Init**（source_file）：all = 证据：`src/agentassay/attribution/__init__.py`
- **Init**（source_file）：all = "cli" 证据：`src/agentassay/cli/__init__.py`
- **Security: Validate port range**（source_file）：Security: Validate port range ⋮---- import streamlit noqa: F401 ⋮---- dashboard pkg = files "agentassay.dashboard" app path = str dashboard pkg.joinpath "app.py" ⋮---- app path = str Path file .resolve .parent.parent / "dashboard" / "app.py" ⋮---- cmd = 证据：`src/agentassay/cli/cmd_dashboard.py`
- **Extract pass/fail**（source_file）：def report command results: str, output: str - None ⋮---- results data = load json results, "results" ⋮---- Extract pass/fail ⋮---- passed list = extract passed list results data ⋮---- passed list = n = len passed list k = sum passed list rate = k / n if n 0 else 0.0 Compute CI if we have data ci lower = 0.0 ci upper = 1.0 ⋮---- ci = wilson interval k, n ci lower = ci.lower ci upper = ci.upper Generate HTML report timestamp = datetime.now timezone.utc .strftime "%Y-%m-%d %H:%M:%S UTC" html = build html n, k, rate, ci lower, ci upper, timestamp out path = Path output ⋮---- Print summary summary = Table title="Report Summary", show header=True ⋮---- verdict class = "pass" if rate = 0.80 else… 证据：`src/agentassay/cli/cmd_report.py`
- **Map YAML keys to AssayConfig fields**（source_file）：assay kwargs: dict str, Any = {} ⋮---- cfg data = load yaml config, "config" ⋮---- Map YAML keys to AssayConfig fields key map = { ⋮---- assay cfg = AssayConfig assay kwargs ⋮---- config table = Table title="Assay Configuration", show header=True ⋮---- scenario data = load yaml scenario, "scenario" ⋮---- If output requested, write the config as JSON for reference ⋮---- output data = { 证据：`src/agentassay/cli/cmd_run.py`
- **Show pytest output**（source_file）：console = Console ⋮---- pass rate = passed / total 100.0 if total 0 else 0.0 ⋮---- ci = wilson interval successes=passed, n=total, confidence=0.95 bar fill class = html = f""" ⋮---- stats = module breakdown module name mod passed = stats "passed" mod failed = stats "failed" mod total = mod passed + mod failed mod rate = mod passed / mod total 100.0 if mod total 0 else 0.0 color = " 3fb950" if mod rate = 80 else " bb8009" if mod rate = 50 else " f85149" ⋮---- def test report command output: str, no browser: bool - None ⋮---- junit xml path = Path tf.name ⋮---- result = subprocess.run Show pytest output ⋮---- tree = ET.parse junit xml path root = tree.getroot testsuite = root.find "testsuite"… 证据：`src/agentassay/cli/cmd_test_report.py`
- **Main**（source_file）：def version message - str ⋮---- def cli - None 证据：`src/agentassay/cli/main.py`
- **Init**（source_file）：all = 证据：`src/agentassay/contracts/__init__.py`
- **-- Internal: scoring ---------------------------------------------------**（source_file）：logger = logging.getLogger name HAS AGENTASSERT = False ⋮---- HAS AGENTASSERT = True ⋮---- class ContractOracle ⋮---- slots = " contract data", " contract name", " constraints" ⋮---- contract = self. contract data "contract" ⋮---- @property def contract name self - str ⋮---- @property def num constraints self - int def evaluate self, trace: ExecutionTrace - ContractEvaluation ⋮---- context = build trace context trace violations: list ContractViolation = ⋮---- c name: str = constraint "name" c type: str = constraint "type" c severity: str = constraint "severity" c condition: str = constraint "condition" ⋮---- score = self. compute score violations has hard violations = any v.severity == "har… 证据：`src/agentassay/contracts/oracle.py`
- **Init**（source_file）：all = 证据：`src/agentassay/core/__init__.py`
- **Init**（source_file）：all = 证据：`src/agentassay/coverage/__init__.py`
- **── Persistence layer ──────────────────────────────────────────────────**（source_file）：DARK CSS = """ LIGHT CSS = """ current theme = st.get option "theme.base" or "dark" ⋮---- page = st.radio ⋮---- ── Persistence layer ────────────────────────────────────────────────── from agentassay.persistence import QueryAPI, ResultStore noqa: E402 - after st.set page config db path = os.environ.get store = ResultStore db path=db path query api = QueryAPI store 证据：`src/agentassay/dashboard/app.py`
- **Init**（source_file）：all = 证据：`src/agentassay/efficiency/__init__.py`
- **Step 1: Extract fingerprints**（source_file）：class BudgetEstimate BaseModel ⋮---- model config = ConfigDict frozen=True behavioral variance: float = Field ge=0.0 recommended n: int = Field ge=2 estimated cost usd: float = Field ge=0.0 variance classification: str savings vs fixed 100: float confidence level: float = Field gt=0.0, lt=1.0 power level: float = Field gt=0.0, lt=1.0 per trial cost usd: float = Field ge=0.0 diagnostics: dict str, float = Field default factory=dict calibration size: int = Field ge=2 VARIANCE THRESHOLDS: dict str, float = { def classify variance variance: float - str class AdaptiveBudgetOptimizer ⋮---- slots = " alpha", " beta", " delta", " min trials", " max trials" ⋮---- @property def alpha self - float ⋮--… 证据：`src/agentassay/efficiency/budget.py`
- **-- Vectorize --**（source_file）：logger = logging.getLogger name DIMENSION NAMES: list str = class BehavioralFingerprint BaseModel ⋮---- model config = ConfigDict frozen=True tool distribution: dict str, float = Field tool entropy: float = Field tool count: int = Field ge=0, description="Total tool calls in this trace." unique tools: int = Field ge=0, description="Distinct tools used." step count: int = Field ge=0, description="Total steps in the trace." max chain depth: int = Field backtrack count: int = Field ge=0, description="Number of retries or corrections detected." output length: int = Field ge=0, description="Character count of final output." output structure hash: str = Field has structured output: bool = Field t… 证据：`src/agentassay/efficiency/fingerprint.py`
- **------------------------------------------------------------------**（source_file）：class MultiFidelityTester ⋮---- slots = " proxy model", " target model", " corr threshold", " estimated corr" ⋮---- @property def proxy model self - str ⋮---- """Name of the cheap proxy model.""" ⋮---- @property def target model self - str ⋮---- """Name of the expensive target model.""" ⋮---- @property def estimated correlation self - float None ⋮---- """Last estimated correlation, or None if not yet computed.""" ⋮---- ------------------------------------------------------------------ Correlation estimation ⋮---- """Estimate behavioral correlation between proxy and target. Uses behavioral fingerprints not raw outputs to compute the Pearson correlation between the proxy and target mean vecto… 证据：`src/agentassay/efficiency/multi_fidelity.py`
- **Regression**（source_file）：baseline fps = BehavioralFingerprint.from trace t for t in baseline traces candidate fps = BehavioralFingerprint.from trace t for t in candidate traces baseline dist = FingerprintDistribution baseline fps candidate dist = FingerprintDistribution candidate fps result = baseline dist.regression test candidate dist, alpha=alpha 证据：`src/agentassay/efficiency/regression.py`
- **Load traces up to limit — touch each in the LRU index**（source_file）：class TraceEncoder json.JSONEncoder ⋮---- def default self, obj: Any - Any class TraceStore ⋮---- slots = " store path", " traces dir", " index path", " index", " max index size" ⋮---- trace id = trace.trace id extra meta = metadata or {} trace data = trace.model dump mode="json" ⋮---- trace path = self. traces dir / f"{trace id}.json.gz" json bytes = json.dumps trace data, cls= TraceEncoder .encode "utf-8" ⋮---- agent id = extra meta.get "agent id", trace.metadata.get "agent id", "unknown" ⋮---- matching ids = self. filter index ⋮---- Load traces up to limit — touch each in the LRU index traces: list ExecutionTrace = ⋮---- trace = self. load trace trace id ⋮---- LRU touch: move accessed en… 证据：`src/agentassay/efficiency/trace_store.py`
- **Wald boundaries same as standard SPRT**（source_file）：class WarmStartSPRT ⋮---- slots = ⋮---- Wald boundaries same as standard SPRT ⋮---- Per-trial LLR increments ⋮---- Apply prior decay effective prior trials = prior trials effective prior successes = prior successes ⋮---- decay factor = math.exp -prior decay days since prior effective prior trials = int round prior trials decay factor effective prior successes = int round prior successes decay factor Ensure consistency effective prior successes = min effective prior successes, effective prior trials ⋮---- Initialize LLR from prior the warm start prior failures = effective prior trials - effective prior successes ⋮---- State ⋮---- @property def is decided self - bool ⋮---- @property def decis… 证据：`src/agentassay/efficiency/warm_start.py`
- **Resolve aliases**（source_file）：ADAPTER REGISTRY: dict str, type AgentAdapter = { ALIASES: dict str, str = { ⋮---- normalized = framework.strip .lower .replace " ", "" Resolve aliases resolved = ALIASES.get normalized, normalized Look up adapter class adapter cls = ADAPTER REGISTRY.get resolved ⋮---- supported = sorted ADAPTER REGISTRY.keys ⋮---- Special handling for custom adapter: first positional arg is callable fn ⋮---- def list adapters - dict str, type AgentAdapter def list aliases - dict str, str all = 证据：`src/agentassay/integrations/__init__.py`
- **AutoGen v0.4 uses async; try run sync first**（source_file）：logger = logging.getLogger name INSTALL HINT = def check autogen installed - None class AutoGenAdapter AgentAdapter ⋮---- framework: str = "autogen" ⋮---- resolved model = model ⋮---- llm config = getattr agent, "llm config", None ⋮---- config list = llm config.get "config list", ⋮---- resolved model = config list 0 .get "model", "unknown" ⋮---- resolved model = llm config "model" resolved name = agent name ⋮---- resolved name = getattr agent, "name", None or "autogen-agent" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- user message = self. build mes… 证据：`src/agentassay/integrations/autogen_adapter.py`
- **Base**（source_file）：class AdapterError Exception class FrameworkNotInstalledError ImportError class AgentAdapter ABC ⋮---- framework: str ⋮---- @abstractmethod def run self, input data: dict str, Any - ExecutionTrace ⋮---- @abstractmethod def to callable self - Callable dict str, Any , ExecutionTrace def get config self - AgentConfig ⋮---- @property def model self - str ⋮---- @property def agent name self - str def repr self - str 证据：`src/agentassay/integrations/base.py`
- **-- Rationale agent reasoning ------------------------------------**（source_file）：logger = logging.getLogger name INSTALL HINT = def check boto3 installed - None class BedrockAgentsAdapter AgentAdapter ⋮---- framework: str = "bedrock" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 session id = self. session id or str uuid.uuid4 overall start = time.perf counter ⋮---- client = self. get client user input = self. build user input input data response = client.invoke agent ⋮---- total ms = time.perf counter - overall start 1000.0 ⋮---- error msg = f"{type exc . name }: {exc}" ⋮---- def to callable self - Callable dict str, Any , ExecutionTrace def get config self - AgentConf… 证据：`src/agentassay/integrations/bedrock_adapter.py`
- **Crewai Adapter**（source_file）：logger = logging.getLogger name INSTALL HINT = "CrewAI adapter requires crewai. Install with: pip install agentassay crewai " def check crewai installed - None class CrewAIAdapter AgentAdapter ⋮---- framework: str = "crewai" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- kickoff start = time.perf counter result = self. crew.kickoff inputs=input data kickoff duration ms = time.perf counter - kickoff start 1000.0 steps = self. extract steps result, kickoff duration ms output = self. extract output result total ms = time.perf counter - overall start 1000… 证据：`src/agentassay/integrations/crewai_adapter.py`
- **Langgraph Adapter**（source_file）：logger = logging.getLogger name INSTALL HINT = def check langgraph installed - None class LangGraphAdapter AgentAdapter ⋮---- framework: str = "langgraph" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- total ms = time.perf counter - overall start 1000.0 ⋮---- error msg = f"{type exc . name }: {exc}" ⋮---- def to callable self - Callable dict str, Any , ExecutionTrace def run stream self, input data: dict str, Any - tuple list StepTrace , Any ⋮---- steps: list StepTrace = last output: Any = None step index = 0 ⋮---- step start = time.perf counter ⋮----… 证据：`src/agentassay/integrations/langgraph_adapter.py`
- **Mcp Adapter**（source_file）：logger = logging.getLogger name INSTALL HINT = def check mcp installed - None class MCPToolsAdapter AgentAdapter ⋮---- framework: str = "mcp" ⋮---- meta = metadata or {} ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- total ms = time.perf counter - overall start 1000.0 ⋮---- error msg = f"{type exc . name }: {exc}" ⋮---- def to callable self - Callable dict str, Any , ExecutionTrace def get config self - AgentConfig ⋮---- @staticmethod def detect mode client: Any - str ⋮---- client type = type client . name .lower ⋮---- @staticmethod def build user inp… 证据：`src/agentassay/integrations/mcp_adapter.py`
- **Openai Adapter**（source_file）：logger = logging.getLogger name INSTALL HINT = def check openai agents installed - None class OpenAIAgentsAdapter AgentAdapter ⋮---- framework: str = "openai" ⋮---- resolved model = model ⋮---- resolved model = getattr agent, "model", "unknown" or "unknown" resolved name = agent name ⋮---- resolved name = getattr agent, "name", None or "openai-agent" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- user input = self. build user input input data run start = time.perf counter result = Runner.run sync self. agent, user input run duration ms = time.perf cou… 证据：`src/agentassay/integrations/openai_adapter.py`
- **Extract result value as string**（source_file）：logger = logging.getLogger name INSTALL HINT = def check semantic kernel installed - None class SemanticKernelAdapter AgentAdapter ⋮---- framework: str = "semantic kernel" ⋮---- def run self, input data: dict str, Any - ExecutionTrace ⋮---- scenario id = input data.get "scenario id", "default" trace id = str uuid.uuid4 overall start = time.perf counter ⋮---- total ms = time.perf counter - overall start 1000.0 ⋮---- error msg = f"{type exc . name }: {exc}" ⋮---- def to callable self - Callable dict str, Any , ExecutionTrace def get config self ⋮---- merged metadata = { ⋮---- def invoke kernel self, input data: dict str, Any - tuple list StepTrace , Any, float ⋮---- kernel args = self. build… 证据：`src/agentassay/integrations/semantic_kernel_adapter.py`
- 其余 19 条证据见 `AI_CONTEXT_PACK.json` 或 `EVIDENCE_INDEX.json`。

## 宿主 AI 必须遵守的规则

- **把本资产当作开工前上下文，而不是运行环境。**：AI Context Pack 只包含证据化项目理解，不包含目标项目的可执行状态。 证据：`README.md`, `examples/README.md`, `experiments/README.md`
- **回答用户时区分可预览内容与必须安装后才能验证的内容。**：安装前体验的消费者价值来自降低误装和误判，而不是伪装成真实运行。 证据：`README.md`, `examples/README.md`, `experiments/README.md`

## 用户开工前应该回答的问题

- 你准备在哪个宿主 AI 或本地环境中使用它？
- 你只是想先体验工作流，还是准备真实安装？
- 你最在意的是安装成本、输出质量、还是和现有规则的冲突？

## 验收标准

- 所有能力声明都能回指到 evidence_refs 中的文件路径。
- AI_CONTEXT_PACK.md 没有把预览包装成真实运行。
- 用户能在 3 分钟内看懂适合谁、能做什么、如何开始和风险边界。

---

## Doramagic Context Augmentation

下面内容用于强化 Repomix/AI Context Pack 主体。Human Manual 只提供阅读骨架；踩坑日志会被转成宿主 AI 必须遵守的工作约束。

## Human Manual 骨架

使用规则：这里只是项目阅读路线和显著性信号，不是事实权威。具体事实仍必须回到 repo evidence / Claim Graph。

宿主 AI 硬性规则：
- 不得把页标题、章节顺序、摘要或 importance 当作项目事实证据。
- 解释 Human Manual 骨架时，必须明确说它只是阅读路线/显著性信号。
- 能力、安装、兼容性、运行状态和风险判断必须引用 repo evidence、source path 或 Claim Graph。

- **Project Overview & Six-Layer Architecture**：importance `high`
  - source_paths: README.md, docs/architecture/overview.md, src/agentassay/core/models.py, src/agentassay/core/runner.py, src/agentassay/__init__.py
- **Token-Efficient Engine: Fingerprints, Adaptive Budgets & Statistical Verdicts**：importance `high`
  - source_paths: src/agentassay/efficiency/fingerprint.py, src/agentassay/efficiency/budget.py, src/agentassay/efficiency/multi_fidelity.py, src/agentassay/efficiency/warm_start.py, src/agentassay/efficiency/trace_store.py
- **Framework Adapters, CLI, pytest Plugin & Dashboard**：importance `high`
  - source_paths: src/agentassay/integrations/base.py, src/agentassay/integrations/langgraph_adapter.py, src/agentassay/integrations/crewai_adapter.py, src/agentassay/integrations/autogen_adapter.py, src/agentassay/integrations/openai_adapter.py
- **Analysis, Mutation, Metamorphic Testing & Deployment Gates**：importance `medium`
  - source_paths: src/agentassay/efficiency/trace_store.py, src/agentassay/mutation/runner.py, src/agentassay/mutation/operators.py, src/agentassay/mutation/prompt_ops.py, src/agentassay/mutation/tool_ops.py

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `7da9a259717aee9bfb0abc40ce57b0e1df425881`
- inspected_files: `pyproject.toml`, `README.md`, `docs/index.md`, `docs/operations/troubleshooting.md`, `docs/architecture/overview.md`, `docs/getting-started/quickstart.md`, `docs/getting-started/installation.md`, `docs/guides/dashboard.md`, `docs/guides/ci-cd-integration.md`, `docs/guides/pytest-plugin.md`, `docs/guides/framework-adapters.md`, `docs/concepts/mutation-testing.md`, `docs/concepts/token-efficient-testing.md`, `docs/concepts/stochastic-testing.md`, `docs/concepts/coverage.md`, `docs/reference/cli.md`, `docs/reference/glossary.md`, `docs/reference/configuration.md`, `docs/guides/adapters/smolagents.md`, `docs/guides/adapters/mcp.md`

宿主 AI 硬性规则：
- 没有 repo_clone_verified=true 时，不得声称已经读过源码。
- 没有 repo_inspection_verified=true 时，不得把 README/docs/package 文件判断写成事实。
- 没有 quick_start_verified=true 时，不得声称 Quick Start 已跑通。

## Doramagic Pitfall Constraints / 踩坑约束

这些规则来自 Doramagic 发现、验证或编译过程中的项目专属坑点。宿主 AI 必须把它们当作工作约束，而不是普通说明文字。

### Constraint 1: 能力判断依赖假设

- Trigger: README/documentation is current enough for a first validation pass.
- Host AI rule: 将假设转成下游验证清单。
- Why it matters: 假设不成立时，用户拿不到承诺的能力。
- Evidence: capability.assumptions | github_repo:1174343741 | https://github.com/qualixar/agentassay | README/documentation is current enough for a first validation pass.
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 2: 维护活跃度未知

- Trigger: 未记录 last_activity_observed。
- Host AI rule: 补 GitHub 最近 commit、release、issue/PR 响应信号。
- Why it matters: 新项目、停更项目和活跃项目会被混在一起，推荐信任度下降。
- Evidence: evidence.maintainer_signals | github_repo:1174343741 | https://github.com/qualixar/agentassay | last_activity_observed missing
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

- Trigger: no_demo
- Evidence: downstream_validation.risk_items | github_repo:1174343741 | https://github.com/qualixar/agentassay | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 4: 存在评分风险

- Trigger: no_demo
- Why it matters: 风险会影响是否适合普通用户安装。
- Evidence: risks.scoring_risks | github_repo:1174343741 | https://github.com/qualixar/agentassay | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 5: issue/PR 响应质量未知

- Trigger: issue_or_pr_quality=unknown。
- Host AI rule: 抽样最近 issue/PR，判断是否长期无人处理。
- Why it matters: 用户无法判断遇到问题后是否有人维护。
- Evidence: evidence.maintainer_signals | github_repo:1174343741 | https://github.com/qualixar/agentassay | issue_or_pr_quality=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 6: 发布节奏不明确

- Trigger: release_recency=unknown。
- Host AI rule: 确认最近 release/tag 和 README 安装命令是否一致。
- Why it matters: 安装命令和文档可能落后于代码，用户踩坑概率升高。
- Evidence: evidence.maintainer_signals | github_repo:1174343741 | https://github.com/qualixar/agentassay | release_recency=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。