v1.4.0 — Public MCP Setup Flow

v1.4.0 — Public MCP Setup Flow

This release makes the first-run MCP setup path safer and friendlier for new users.

Highlights

• Adds python scripts/setup_mcp_tools.py as a check-first default entry point.
• Keeps the recommended path minimal: exam-memory installs with pip install -e . instead of pulling optional embedding/generation extras.
• Reports both installed and configured MCP status so users can see whether .mcp.json is already wired.

Setup Improvements

• Adds --yes for auto-confirming the default recommended setup.
• Adds --no-input for non-interactive status checks.
• Supports local overrides for external tools with CHATMEM_HOME, CHATMEM_MCP, CHATMEM_EXE, and ONEFIND_HOME.
• Checks that exam_memory.server, mcp, and yaml are importable before treating exam-memory as installed.

Packaging And Tests

• Fixes shared/exam_memory/pyproject.toml package mapping and package data for editable installs.
• Adds regression tests for MCP setup detection and prompt parsing.

Verification

• python scripts\check_main_extraction.py --require-branch main
• python -m pytest -q --basetemp=.tmp/pytest-main-extraction (346 passed)
• git diff --check

v1.3.6 — cheatsheets public, zhishou exam materials

变更

• 备忘录（cheatsheets）标记为公开安全，允许进入 main 分支
• 知行兽医考试资料提取到 main（基础科目、预防科目、2024年真题解析）
• 调整 main 提取门控规则（denylist），移除 cheatsheets 限制
• exam_config.md 命名规范化

涉及提交

• 74edf33 chore: allow cheatsheets through main extraction gate
• 6c6ee26 feat: extract zhishou exam materials to main
• d519a97 Merge branch 'pr-8-zhishou-exam'

v1.3.5 — Public-Clean Main Extraction (D2 + D3)

Public-safe changes extracted from dev to main, post public-cleanup verification.

Changes (5 files, 2 commits)

• HANDOFF.example.md — new public session template for preparation handoffs
• prompts/new_session_prompt.md — MCP path + Lite/Portable Mode clarifications
• scripts/check_main_extraction.py — hardened denylist (cheatsheets/sources blocked)
• shared/exam_memory/pyproject.toml — setuptools packages find config
• shared/exam_memory/sources.yaml — ai-lab-notes source config

Commits

• 5610149 feat: update public-safe functional files from dev
• 834dccb feat: add HANDOFF.example.md as public session template

This release supersedes v1.3.0 (which pointed to an earlier extraction with denylist leaks).

v1.3.0 — Exam Memory V2 Modules + Progress Infrastructure

What's New

Exam Memory V2 核心模块

• difficulty_calibrator.py — 难度校准器，基于答题统计动态调整题目难度
• distill_export.py — 知识蒸馏 logprobs 采集与导出
• text_pipeline.py — 文本处理管道（分块、清洗、嵌入）

题库系统增强

• question_bank.py — 扩展 CRUD + LLM 生成管道 + 语义去重
• vector_store.py — numpy-based 语义检索（替代 ChromaDB）
• server.py — MCP 服务端集成

进度追踪基础设施

• shared/progress/ — task-board 模板 + README
• shared/exam_memory/bank/ — 题目 bank 目录框架

Skills 更新

• solve-analyze — WA/TLE 诊断根因 skill
• solve-skeleton — OJ 题骨架模板 skill

测试覆盖

• 新增 6 个测试文件（difficulty_calibrator, distill_export, semantic_dedup, text_pipeline）
• 1700+ 测试用例，全部通过

其他

• 更新 new_session_prompt 以反映 Exam Memory 系统和 ChatMem 集成
• security hardening（path traversal guards, key leak fix）
• .gitignore 更新，忽略运行时生成的 JSON 数据文件

v1.2 — Security Hardening & Code Quality

v1.2 — Security Hardening & Code Quality

Security Fixes

• API key leak fix: Embedding provider abstraction now supports env-var configuration for both local and API-based providers, eliminating hardcoded credentials
• Path traversal guards: Added path validation in server.py and question_bank.py to prevent directory traversal attacks
• Profile schema whitelist: update_user_profile now validates against an allowlist of permitted fields
• vector_store.rebuild(): Now emits an explicit warning before silently clearing the vector store index

Changelog

4 cherry-picked commits from dev branch, addressing Copilot review findings and proactive security hardening.

v1.0.1 — Fix external MCP links

v1.0.1 — 文档修正

修正外部 MCP Server 引用链接，更新为正确的 GitHub 地址：

MCP	旧地址	新地址
ChatMem	github.com/nicobailon/chatmem	github.com/Rimagination/ChatMem/releases
mempalace	无链接	github.com/MemPalace/mempalace
onefind	github.com/...（占位）	github.com/iawnfoanaowt/OneFind

涉及文件

• README.md
• README_CN.md
• docs/mcp-setup-guide.md

三个外部 MCP 均标注为可选，项目本身不依赖它们，所有 Skill 在 MCP 不可用时自动降级为纯本地模式。

v1.0.0 — pass-llm-with-llm

pass-llm-with-llm v1.0.0

AI/算法岗位笔试备考引擎 — Claude Code Skills 驱动的执行框架。

核心特性

• 5 个 Skills：solve-skeleton → solve-analyze → algo-annotation → choice-q-create → choice-q-drill
• 自定义 MCP Server：exam-memory，6 个工具实现跨会话经验持久化
• 11 道真题解答 + ACM/OJ 工具库
• 双通道错误回流：mistake_log.md（快速查阅）+ exam-memory MCP（结构化持久化）

技术栈

• Python 3.10+（MCP Server）
• Claude Code Skills（Markdown 驱动的工作流引擎）
• 可选：sentence-transformers + numpy（V2 语义检索）

开源准备

• 中英文双版 README
• GitHub 社区模板（Issue / PR）
• SECURITY.md + CHANGELOG.md
• 泛化的 AGENTS.md（移除个人数据）

详见 CHANGELOG.md。