v0.20.1

What's Changed

• docs(AGENTS): add World Model Strategy section by @juice094 in #33
• chore(cleanup): reorganize scattered root-level documents by @juice094 in #34
• chore(gitignore): clean null-byte corruption and add .kimi/ by @juice094 in #35
• docs(README,AGENTS): sync World Model positioning and bidirectional linkage by @juice094 in #36
• feat(v0.19.0): SQLite WAL + devkit_index_health + roadmap overhaul by @juice094 in #37
• feat(v0.19.0): devkit_vault_export + data freedom by @juice094 in #38
• release(v0.20.0): 知识完备性 — Vault 图遍历 / 历史追踪 / 检索质量 / Block 引用 by @juice094 in #39
• chore: bump version to 0.20.0 by @juice094 in #40
• chore(v0.20.1): 技术债清偿 — 文档诚实化 + 版本号统一 + RF-7 合规 by @juice094 in #41
• refactor(registry): use 规范化，RF-7 全部合规 by @juice094 in #42
• docs(AGENTS): 技术债登记簿状态更新 by @juice094 in #43
• ci(release): 添加 Linux 构建 job + 更新 README 下载链接 by @juice094 in #44
• chore: post-release cleanup — remove unused unix import + archive obsolete plans by @juice094 in #45
• ci(release): auto-create GitHub Release on tag push by @juice094 in #46
• chore: archive completed RF-6 unwrap audit plan by @juice094 in #47
• refactor(executor): merge redundant #[cfg] branches for wait_with_timeout by @juice094 in #48
• fix(query): 'repos' keyword now returns all registered repositories by @juice094 in #49
• feat(greptimedb): Phase A — feature gate, config schema, and placeholder module by @juice094 in #50
• chore(scripts): add update_readme.py for release asset metadata automation by @juice094 in #51

Full Changelog: v0.18.0...v0.20.1

devbase v0.20.0 — 知识完备性

devbase v0.20.0 — 知识完备性

主题：从"能存"到"好用"，消除知识库能力缺口。

---

✨ 新功能

Vault 双向链接图遍历（Sprint E/F 延伸）

• devkit_vault_graph 支持 BFS 双向遍历：note_id + depth（1–3）参数
• DB-first 构建：从 SQLite outgoing_links 出发，自动补全 incoming 边
• 知识图谱不再只是静态导出，而是可查询的图结构

Vault Git-based 历史追踪（Sprint E）

• devkit_vault_history（第 67 个 tool）：基于 git2 revwalk 的笔记变更追踪
• 支持 blob 内容比对，返回行级 insertions / deletions
• VaultClient trait 扩展：get_vault_history() 统一接口

混合检索质量监控（Sprint F）

• devkit_search_quality（第 68 个 tool）：返回 HybridSearchMetrics
  • latency_ms — 查询耗时
  • keyword_recall / vector_recall — 各路召回数
  • rrf_overlap — keyword & vector 重叠项
  • keyword_source — "tantivy" 或 "sqlite_fallback"
  • rrf_k — 融合常数

Block 引用支持（Sprint G）

• WikiLink.anchor：支持 [[note#heading]] 与 [[note#^block-id]]
• VaultNote.block_refs：块级引用持久化（JSON metadata，无 schema 迁移）
• Vault 导出时自动检测 broken block refs（heading 锚点不存在时报告）

性能回归基线（Sprint C）

• 新增 #[ignore] 性能回归测试：
  • 1k 文档 keyword search < 200ms
  • 10k 文档 keyword search < 500ms
• ADR-003：Redis 缓存评估完成 → 决策放弃引入，现有 SQLite + Tantivy 栈已足够

---

🏛️ 架构原则

Client-Agnostic Principle（客户端无关原则）

• 核心能力零硬编码客户端路径（移除 .clarity 硬编码、泛化 skill sync 接口）
• 适配层（scripts/claude/）与核心代码严格分离
• 删除 Kimi/Claude 后，devbase 的编排、注册、索引、搜索、同步能力完全独立

---

📊 统计

指标	v0.18.0	v0.20.0	变化
MCP Tools	64	68	+4
Tests	437	451	+14
Workspace Crates	18	19	+1
Schema	34	34	稳定

---

🛠️ 修复与优化

• Clippy 清零：生产代码 field-reassign-with-default 修复
• Rustfmt：8 文件格式统一
• Tantivy 健康评分：devkit_index_health（Beta）持续可用
• SQLite WAL：默认启用，并发安全

---

⚠️ 已知限制

• tree-sitter grammar 编译成本仍维持 15–20s（已 feature-gate，可选关闭）
• Vault 历史依赖用户将 vault/ 目录初始化为 Git 仓库
• Block 引用中的 ^block-id 锚点目前仅解析，不验证存在性

---

📦 安装

# Windows
irm https://raw.githubusercontent.com/juice094/devbase/main/scripts/install.ps1 | iex

# 或下载预编译二进制
wget https://github.com/juice094/devbase/releases/download/v0.20.0/devbase-v0.20.0-x86_64-pc-windows-msvc.exe -O devbase.exe

# Linux
wget https://github.com/juice094/devbase/releases/download/v0.20.0/devbase-v0.20.0-x86_64-unknown-linux-gnu -O devbase
chmod +x devbase

---

🙏 致谢

本版本全部功能由单人维护完成，遵循 Client-Agnostic Principle 与 Local-First 架构红线。

devbase v0.18.0

🎯 ClaudeCode Workflow Integration

New MCP Tools (+4 → 64 total)

Tool	Purpose
devkit_project_brief	Generate Markdown project brief for .claude/CLAUDE.md injection
devkit_impact_analysis	Symbol-level blast-radius analysis (BFS call graph + test heuristic)
devkit_session_export	Export agent sessions to Markdown/JSON
devkit_session_import	Bulk import memories from text with [type] prefix

Core Architecture

• Agent Memory Vector Storage (Schema v34): SQLite BLOB + UDF cosine_similarity — zero LLM runtime dependency
• Tiered Auto-Recall: Semantic → Keyword fallback in Skill Runtime
• TUI Three-State MainView: Tab cycles RepoList → VaultList → Session

Assets

• Windows x86_64: devbase-v0.18.0-x86_64-pc-windows-msvc.exe (23 MB)
• Linux x86_64: devbase-v0.18.0-x86_64-unknown-linux-gnu (22 MB)

Quick Start (ClaudeCode)

./scripts/devbase-claude.ps1 -CaptureOnExit

Full changelog: CHANGELOG.md

v0.16.1

What's Changed

• fix(tests): guard disk-based search test with SEARCH_TEST_LOCK by @juice094 in #12
• build: sccache configuration for faster incremental builds by @juice094 in #13
• refactor(mcp): trait-ize remaining inline crate:: calls in repo.rs by @juice094 in #14
• refactor(mcp): complete trait-ization of remaining MCP tool layer by @juice094 in #15
• refactor(knowledge_engine): extract Config load + prepare_repos, eliminate loop-level I/O by @juice094 in #16
• perf(knowledge_engine): reuse Tantivy writer across daemon batch indexing by @juice094 in #17
• docs: module rustdoc + ADR-004/005 + i18n dead_code clarification by @juice094 in #18
• docs: update project facade files (README ecosystem) by @juice094 in #19
• test(knowledge_engine): 补齐 index.rs 与 index_state.rs 单元测试 by @juice094 in #20
• ci: upgrade actions/checkout v4 to v6 (Node.js 24) by @juice094 in #21
• test(vault): 补齐 indexer.rs 单元测试，修复 id field 分词导致删除失效 by @juice094 in #22
• test(semantic_index): 补齐 persist.rs 单元测试 by @juice094 in #23
• test(discovery_engine): 补齐 package_json、go_mod、similar_projects 测试 by @juice094 in #24
• chore(deps): patch upgrade assert_cmd/blake3/tantivy/tokio by @juice094 in #25
• docs: 对齐 README Tool 矩阵与实际代码（48 tools） by @juice094 in #26
• feat(mcp): add devkit_evaluate — AI self-evaluation tool by @juice094 in #27
• feat(agent-contexts): v0.16.0 AI Agent OS — P1+P2+P3 by @juice094 in #28
• feat(agent-contexts): v0.16.1 Workflow-Session binding by @juice094 in #29

Full Changelog: v0.15.2...v0.16.1

v0.16.0

What's Changed

• fix(tests): guard disk-based search test with SEARCH_TEST_LOCK by @juice094 in #12
• build: sccache configuration for faster incremental builds by @juice094 in #13
• refactor(mcp): trait-ize remaining inline crate:: calls in repo.rs by @juice094 in #14
• refactor(mcp): complete trait-ization of remaining MCP tool layer by @juice094 in #15
• refactor(knowledge_engine): extract Config load + prepare_repos, eliminate loop-level I/O by @juice094 in #16
• perf(knowledge_engine): reuse Tantivy writer across daemon batch indexing by @juice094 in #17
• docs: module rustdoc + ADR-004/005 + i18n dead_code clarification by @juice094 in #18
• docs: update project facade files (README ecosystem) by @juice094 in #19
• test(knowledge_engine): 补齐 index.rs 与 index_state.rs 单元测试 by @juice094 in #20
• ci: upgrade actions/checkout v4 to v6 (Node.js 24) by @juice094 in #21
• test(vault): 补齐 indexer.rs 单元测试，修复 id field 分词导致删除失效 by @juice094 in #22
• test(semantic_index): 补齐 persist.rs 单元测试 by @juice094 in #23
• test(discovery_engine): 补齐 package_json、go_mod、similar_projects 测试 by @juice094 in #24
• chore(deps): patch upgrade assert_cmd/blake3/tantivy/tokio by @juice094 in #25
• docs: 对齐 README Tool 矩阵与实际代码（48 tools） by @juice094 in #26
• feat(mcp): add devkit_evaluate — AI self-evaluation tool by @juice094 in #27
• feat(agent-contexts): v0.16.0 AI Agent OS — P1+P2+P3 by @juice094 in #28
• feat(agent-contexts): v0.16.1 Workflow-Session binding by @juice094 in #29

Full Changelog: v0.15.2...v0.16.0

v0.15.2

What's Changed

• refactor(storage): reduce crate:: refs 19→5 via use consolidation by @juice094 in #11

Full Changelog: v0.15.1...v0.15.2

v0.15.1

What's Changed

• fix(v0.15.1): clippy清零 + feature-flags warnings + crate README补全 by @juice094 in #9
• refactor(daemon): reduce crate:: refs 19→9 via use consolidation by @juice094 in #10

Full Changelog: v0.15.0...v0.15.1

v0.14.3

What's Changed

• fix(tests): eliminate flaky tests via hermetic StorageBackend by @juice094 in #4
• fix(clippy): resolve 7 lints for Rust 1.95 by @juice094 in #5

New Contributors

• @juice094 made their first contribution in #4

Full Changelog: v0.12.0...v0.14.3

v0.12.0 — Phase 2 Completion + Managed-Gate Fail-Safe Sync

What'"'"'s New

Phase 2: Entities Unified Architecture (Complete)

• Schema v22: vault_notes, papers, workflows orphan tables dropped. entities is now the sole source of truth for all entity types.
• Generic upsert_entity abstraction for entity dual-write.
• ENTITY_TYPE_* constants extracted across 10 files.

Managed-Gate Fail-Safe Defaults

• devbase sync now defaults to safe: only repos with management tags are synced.
• Management tags: mirror, reference, third-party, collaborative, team, own-project, tool, active, managed.
• scan --register no longer auto-tags repos with "discovered".
• Untagged repos are tracked but skipped; use --filter-tags for explicit selection.

Other Improvements

• .devbase-ignore for directory-level scan exclusion.
• Sync output now shows count of skipped unmanaged repos.
• Flaky search tests fixed (eliminated set_var race conditions).
• Integration tests hardened: 11 passed (was 9).

⚠️ Breaking Change

Repos previously tagged "discovered" are no longer synced by default.

Migration:

devbase tag <repo-id> managed

Or use any management tag (mirror, active, etc.) to opt the repo into sync.

Stats

• 379 unit tests passed
• 11 integration tests passed
• 0 clippy warnings

v0.10.0 — AI gains self-awareness of its own limits

devbase v0.10.0 Release Notes

Release Date: 2026-04-26
Schema Version: v19
Tests: 288 passed / 0 failed / 3 ignored

---

Highlights

L3 Risk Layer MVP — known_limits

系统首次具备自我边界意识。known_limits 表记录 hard vetoes、已知缺陷和外部依赖风险，并在 Skill 执行前自动审计。

# 查看当前系统约束
devbase limit list

# 解决一个已知限制（自动记录到 L4 元认知层）
devbase limit resolve my-limit --reason "已验证安全"

# 从 AGENTS.md 自动填充 hard vetoes
devbase limit seed

L4 元认知层 MVP — knowledge_meta

记录人类对 L1-L3 的纠正和反馈，形成认知纠错闭环。

# resolve 时带 --reason 会自动创建 L4 记录
devbase limit resolve hard-veto-xxx --reason "经人工复核，该限制在容器环境中可豁免"

Hard Veto 运行时守卫

Skill 执行前自动检查未解决的 hard veto。发现时：不阻止执行，但在 stderr 头部注入 [HARD-VETO-WARNING] 警告，同时写入 OpLog 审计。

[HARD-VETO-WARNING] Skill 'embed-repo' executed with 5 unresolved hard veto(s):
- [hard-veto-xxx] 禁止闭源 / 云端强制 / 数据外泄
- [hard-veto-yyy] 禁止 Docker / RAG(Qdrant) / GUI(Electron)
...

---

Schema Changes

版本	表	说明
v18	known_limits	L3 风险层：id, category, description, source, severity, first_seen_at, last_checked_at, mitigated
v19	knowledge_meta	L4 元认知层：id, target_level, target_id, correction_type, correction_json, confidence, created_at

---

New MCP Tools

Tool	Tier	说明
devkit_known_limit_store	Beta	存储/更新 known limit
devkit_known_limit_list	Beta	列出 known limits（支持 category/mitigated 过滤）

MCP tool 总数：37

---

New CLI Commands

devbase limit add <id> --category <cat> --description <desc> [--source <src>] [--severity <1-5>]
devbase limit list [--category <cat>] [--mitigated] [--json]
devbase limit resolve <id> [--reason <reason>]
devbase limit delete <id>
devbase limit seed

---

Migration Notes

从 v0.9.0 升级时，registry 会自动备份并迁移到 Schema v19（knowledge_meta 表）。无需手动操作。

---

Known Limitations

• Hard Veto 守卫当前为警告模式（不阻止执行），未来可能支持 --block-on-veto 配置
• knowledge_meta 尚未暴露 MCP tool，仅通过 CLI limit resolve --reason 间接写入
• L0-L2 知识层已存在对应表（entities/skills/vault_notes），但尚未显式标记 knowledge_level