v0.8.2

Full Changelog: v0.8.1...v0.8.2

v0.8.1

What's Changed

• Assert prompt framing at the trace boundary by @coderabbitai[bot] in #59
• Enforce capability policies at tool dispatch by @coderabbitai[bot] in #60

Full Changelog: v0.8.0...v0.8.1

v0.8.0

What's Changed

• fix(memory): stop duplicate memories and self-referencing conflict pairs by @TQSY114514 in #39
• fix(memory): enforce deduplication at the database write layer by @TQSY114514 in #40
• feat(evolution): bounded strategy library with LLM reflection (Hermes-style) by @TQSY114514 in #41
• feat: P0 agent loop reliability - typed-hash loop guard, token-budgeted compaction, overflow self-heal, session-scoped permissions by @TQSY114514 in #42
• chore(deps)(deps): bump the production-dependencies group across 1 directory with 3 updates by @dependabot[bot] in #45
• chore(deps-dev)(deps-dev): bump the dev-dependencies group across 1 directory with 4 updates by @dependabot[bot] in #47
• feat: brief-plan wave 1 — staged tool routing, shrink-retry, TUI approvals, README workflow by @TQSY114514 in #48
• feat: Brief plan 60-day items — project memory scoping, in-chat cost visibility, onboarding choices & external config import, Docker safe default by @TQSY114514 in #50
• feat: Brief plan 90-day items — skill permissions, Arena metrics board, security docs, release checksums, public coding eval by @TQSY114514 in #51
• feat: positioning overhaul — Local-first Agent Workbench identity, honest self-radar by @TQSY114514 in #52
• feat: router-compare harness — measure staged tool-routing impact by @TQSY114514 in #53
• docs: 主流 Agent 工具竞品调研与雷达图对比 by @TQSY114514 in #49
• feat: surface capability-axis ask reasons in permission prompts (P0-2 polish) by @TQSY114514 in #54
• fix(security): resolve CodeRabbit Security findings TQS-5/6/7 by @TQSY114514 in #55
• feat: Agent 核心交互重构、输入框吸顶 HUD 与实用功能增强 by @TQSY114514 in #56
• feat: 全链路打通思考与正文逐字流式输出、任务计划待办预览、多子代理并行矩阵与Windows任务栏Logo修复 by @TQSY114514 in #57
• feat: 打通真·逐字流式输出、气泡内工具卡片归位与输入框吸顶任务甲板重构 by @TQSY114514 in #58

Full Changelog: v0.7.5...v0.8.0

v0.7.5

What's Changed

• feat(ux): P0/P1 UX polish — Ctrl+E, Undo/Redo, stop feedback, command palette, toasts by @TQSY114514 in #21
• Remove hardcoded secret in app/test/tui/session.test.js by @coderabbitai[bot] in #34
• Remove hardcoded secret in app/test/providerMask.test.js by @coderabbitai[bot] in #32
• Remove hardcoded secret in app/test/toolResultMiddleware.test.js by @coderabbitai[bot] in #30
• Bump nanoid to 3.3.18 for advisory by @coderabbitai[bot] in #33
• chore(deps)(deps): bump the production-dependencies group across 1 directory with 3 updates by @dependabot[bot] in #18
• chore(deps)(deps): bump the production-dependencies group in /app/tui with 2 updates by @dependabot[bot] in #17
• Remove hardcoded secret in app/test/toolResultMiddleware.test.js by @coderabbitai[bot] in #35
• fix: desktop agent UX polish — sidebar, credits, i18n, permissions, status bar by @TQSY114514 in #36
• fix: AgentSettings i18n + 5 code quality fixes by @TQSY114514 in #37
• feat: Agent Core 全面增强 — 竞品调研 + P0/P1/P2 全部完成 by @TQSY114514 in #38

New Contributors

• @coderabbitai[bot] made their first contribution in #34
• @dependabot[bot] made their first contribution in #18

Full Changelog: v0.7.4...v0.7.5

v0.7.4

Full Changelog: v0.7.3...v0.7.4

v0.7.3

Full Changelog: v0.7.2...v0.7.3

v0.7.2

Full Changelog: v0.7.1...v0.7.2

v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

Full Changelog: v0.7.0...v0.7.1

v0.7.0

Full Changelog: v0.6.0...v0.7.0

v0.6.0 — 竞品对标全面优化

v0.6.0 — 竞品对标全面优化

本次发布一次性完成 4 个优化 Phase（Phase 1-4），覆盖稳定基础、Agent 竞争力、平台扩展与差异化深耕。核心模块新增 16 个测试文件，总计 307 个测试全部通过。

Phase 1 · 稳定基础

• Token 精确计算：引入 js-tiktoken（cl100k_base）替换字符估算，compaction 触发时机更准确
• DB migration 机制：PRAGMA table_info 检查列 + schema_version 增量迁移，失败显式告警
• Sandbox 白名单加固：参数组合检查，拦截 python -c/node -e/npx 等动态执行与 git 危险命令
• better-sqlite3 事务优化：arena_vote / model_score / 记忆批量 / 会话删除级联使用 db.transaction()

Phase 2 · Agent 竞争力

• Repo Map：文件树 + 顶层 symbol 生成，tool loop 注入，git diff 增量更新
• Git 自动 commit 独立化 + /undo：写文件后即时 commit，checkpoint 与 git 打通
• Lint/Test 自动修复循环：文件变更后自动跑 lint/test，错误注入上下文最多重试 3 轮
• credentialPool 指数退避：30s → 60s → 120s → 600s，/models 验证恢复，401 细分永久/临时

Phase 3 · 平台扩展

• CLI/Headless 模式：agentCore 独立于 Electron，aether "..." --model + --json 输出
• MCP Server 市场：从 Registry API 拉取索引，一键安装热门 server（IPC 三件套就绪）
• Arena ELO → modelRouter 联动：suggestModelForTier 加权 ELO，Auto 模式自动选模型
• 统一搜索：messages + memories + files 统一入口，分类展示，FTS5 全文检索

Phase 4 · 差异化深耕

• Knowledge Graph 增强：二度邻居查询、Graph visualization、图驱动 context injection
• Skill 自动创建增强：参数模板记录，生成带参数 SKILL.md，与 habitLearner 打通
• Scheduled Agent：cron 调度器 + 每日审查 / 依赖检查 / 备份定时任务
• Anthropic 流式 tool_use 完整状态机：content_block_start → delta → stop，参数不丢失
• Prompt Injection 深度防护：外部内容剥离指令，缓存保留 external 标记，内容截断

测试

新增 16 个测试文件，307 个测试全部通过，覆盖 sandbox、toolLoop、credentialPool、modelRouter、adapters、knowledgeGraph、skillSelfCreate、promptInjection 等核心模块。

Full Changelog: v0.5.24...v0.6.0