Skip to content

Releases

v0.3.0 — Code Quality Overhaul + UI Redesign

v0.3.0 — Code Quality Overhaul + UI Redesign Latest
Latest

Choose a tag to compare

View all tags
@Epiphany-Leon Epiphany-Leon released this 06 Sep 16:02
v0.3.0
a35d86f

🚀 v0.3.0 — Code Quality Overhaul + UI Redesign

Massive quality upgrade: 41 files changed, +6,787 / -2,324 lines. Zero functional regressions.

✨ UI Redesign

Tab Architecture: 6 → 3

  • 🔨 锻造 — Forge list / Darwin optimization / Co-evolution
  • 🔗 融合 — Taotie fusion / Skill orchestration
  • 🌙 闲时 — Dreaming console / history / health report

Design System Upgrade:

  • Pill-style main tabs with brand-color active state + subtle shadow
  • Section tabs for inner navigation
  • Card system: 10px radius, 1px border, hover lift + soft shadow
  • Unified buttons: 30px height, 8px radius, primary with shadow
  • Pill-shaped status tags (20px radius)
  • Gradient quality bars with smooth transitions
  • Animated timelines with cleaner dot layout
  • Custom scrollbar styling
  • Fade-in animations on panel load
  • All colors use CSS variables — fully DSH theme-aware

🔧 Code Quality

Type Safety:

  • 54.7% reduction in any types (~316 → 143)
  • 10+ new interface types for service dependencies
  • ForgeOrchestrator: 0 any (down from 18) with 7 typed dep interfaces
  • BaseAgent: typed LLM response + 4 safe accessor methods
  • Client: ~40+ any → 6 (all legitimate: API responses, CSS custom props, catch errors)
  • UnknownRecord pattern for backend data access

New Utilities Layer:

  • src/utils/constants.ts — 30+ magic numbers centralized
  • src/utils/helpers.ts — safe JSON parsing, unified HTTP responses, ID generation, etc.
  • src/types/api.ts — API handler and response types

ForgePanel Refactor (2,831 → 3,229 lines, better structure):

  • 12 extracted inline sub-components
  • All event handlers wrapped with useCallback
  • Clear section boundaries with comment headers
  • 3 main view branches: list / run detail / skill detail

Performance:

  • Active state checks: O(n) → O(1) with Set
  • Safe JSON parsing with multi-strategy fallback
  • Polling cleanup verified for all intervals
  • Memoized handlers to prevent unnecessary re-renders

Error Handling:

  • Unified HTTP response format: { ok: false, error: string, code?: string }
  • safeParseJSON with try/catch + silent fallback
  • Boundary checks for null/undefined/empty arrays

📚 Documentation & Infra

  • CHANGELOG.md: v0.2.0 + v0.3.0 in Keep a Changelog format
  • README: Updated features, architecture, roadmap (Phase 5+6 done)
  • Configuration docs: 14 → 50+ config items with grouping
  • FAQ: New sections on evolution modes + dreaming
  • .editorconfig: New file for consistent formatting
  • package.json: v0.3.0, added dev script, updated keywords
  • tsconfig.json: Added forceConsistentCasingInFileNames
  • .gitignore: Expanded with more patterns

📊 Stats

  • 47 TypeScript/TSX files
  • 30,005 total lines of code
  • 12 core service modules
  • 6 LLM agents
  • ~80 HTTP API endpoints
  • 50+ configuration items
  • 0 TypeScript strict errors

Built with ❤️ for the DSH community

Assets 2
Loading

v0.2.0 — Phase 6: CoEvo / Dreaming / Orchestrator

v0.2.0 — Phase 6: CoEvo / Dreaming / Orchestrator

Choose a tag to compare

View all tags
@Epiphany-Leon Epiphany-Leon released this 06 Sep 14:48
v0.2.0
bbc5c62

🚀 v0.2.0 — Phase 6: Multi-Agent Collaboration Deepening

Massive update with 3 major new capabilities and full UI support.

✨ New Features

🧬 Co-Evolution Mode (CoEvo)

  • AdversarialTestGenerator — 6 types of adversarial tests: edge_case, counterexample, fuzz, security_boundary, cross_scenario, contradictory
  • Double-loop evolution — Odd rounds: skill evolves; Even rounds: test suite evolves
  • Test suite strength evaluation — Defect detection rate + type diversity + dimension coverage + difficulty distribution
  • Weak case pruning — Auto-prune tests with too-high pass rates to prevent bloat
  • Human-in-the-loop — Pause each round for approval
  • 7 HTTP API endpoints — start/status/approve/reject/stop/runs

🌙 Dreaming Mode

  • Skill library health check — Quality distribution, 30-day usage, zombie skill ratio, duplication rate
  • Batch Darwin optimization — Auto-optimize low-scoring skills, prioritized by usage frequency
  • Auto skill fusion — Auto-detect similar skill groups and start Taotie fusion
  • Improvement suggestions — Heuristic-based, zero LLM cost
  • 3 trigger modes — Manual / Scheduled cron / Idle-triggered
  • Safety guarantees — Never auto-delete, strictly non-regressive, always interruptible
  • 5 HTTP API endpoints

🎯 Skill Orchestrator

  • TaskDecomposer — 3 modes: fast (heuristic) / llm (deep) / auto (adaptive)
  • WorkflowBuilder — Skill matching + topological sort + cycle detection
  • DAG execution order — Based on dependsOn relationships
  • Dynamic skill registration — Auto-register workflow-required skills via SkillRegistry
  • Markdown execution guide generation
  • DSH tool: orchestrate_skills
  • 2 HTTP API endpoints

🎨 UI Improvements

  • 4 new sub-tabs in Forge sidebar: 🔨 Forge / 🧬 Optimize / 🔗 Fusion / ⚔️ Co-Evolution / 🌙 Dreaming / 🎯 Orchestrate
  • Settings gear icon in sidebar header — quick settings panel
  • Click-outside-to-close sidebar — default on
  • Squeeze mode (experimental) — sidebar pushes main content instead of floating
  • Settings page overhaul — HTTP API based, 9 collapsible sections, feature overview cards
  • All new modes have full UI — status cards, progress indicators, detail views, timelines

📊 Stats

  • 21 files changed, +13,125 lines
  • 10 core service modules in total
  • ~80 HTTP API endpoints
  • Host bundle: 366KB (gzip: 94KB)
Loading

v0.1.0 — First Release / 首个正式发布

v0.1.0 — First Release / 首个正式发布

Choose a tag to compare

View all tags
@Epiphany-Leon Epiphany-Leon released this 06 Sep 08:04
v0.1.0
5c6ea02

v0.1.0

First official release 🎉

✨ Core Features

🔨 7-Layer Quality Gate Forging Pipeline

  • Trigger → Extract → Generate → Verify → Iterate → Audit → Approve
  • Structured quality control at every stage
  • Failures are traceable, retryable, and never silent

🤖 Multi-Agent Collaborative Forging

  • ExtractorAgent: Distills methodology from conversations
  • GeneratorAgent: Produces complete SKILL.md documents
  • VerifierAgent: 5-dimension quality assessment with test cases
  • RefinerAgent: Targeted optimization on low-score dimensions
  • SecurityAuditor: Static security scanning

🔄 Verification-Driven Iteration

  • Incremental optimization, not full rewrites
  • Strict no-regression: new version score must >= old version
  • Early stop after 2 consecutive rounds with no improvement
  • Complete iteration history tracking

🛡️ 4-Layer Security Defense

  • Minimum privilege during generation
  • Sandbox isolation during verification
  • Version audit on入库
  • Token budget during usage

🔗 Full Skill Lineage Tracking

  • Trace every skill back to its source conversation
  • Version history with one-click rollback
  • Dependency graph and related skill recommendations

🎯 Intent-Aware Smart Injection

  • Analyzes task intent before recalling skills
  • 4-dimensional weighted scoring + token budget management
  • Reward-driven: frequently useful skills rank higher

📊 Incremental Experience Accumulation

  • Accumulates experience traces every conversation turn
  • Auto-triggers forging when experience density reaches threshold
  • Idle trigger as fallback

🎨 Native DSH Web UI Integration

  • Right sidebar (File Explorer / Forge / Stats)
  • Global top bar injection
  • Toast notifications + in-conversation approval
  • Statistics dashboard (7-day trend / Top 5 / injection stats)

🛠 Tech Stack

  • TypeScript strict mode (0 errors)
  • Cordis plugin architecture
  • Zod / Schemastery schema validation
  • React 18 + CSS Variables
  • Zero additional runtime dependencies

📦 Installation

dsh plugin --profile web add dsh-skill-forge

🙏 Acknowledgments

Inspired by: SkillOpt / EvoSkills / Darwin.skill / Taotie.skill / MemoraX Code

v0.1.0 中文版

首个正式发布版本 🎉

✨ 核心特性

🔨 七层质量门锻造流水线

  • 触发 → 提取 → 生成 → 验证 → 迭代 → 审计 → 批准
  • 每道关卡都有结构化质量门控
  • 失败可追溯、可重试、从不静默消失

🤖 多 Agent 协作锻造

  • ExtractorAgent：从对话中提炼方法论骨架
  • GeneratorAgent：生成完整规范的 SKILL.md
  • VerifierAgent：5 维度质量评估 + 测试用例验证
  • RefinerAgent：针对低分维度精准优化
  • SecurityAuditor：静态安全扫描

🔄 验证驱动迭代优化

  • 增量式优化，不是推倒重写
  • 严格不回退：新版本分数必须 >= 旧版本
  • 连续两轮无进步自动早停
  • 完整迭代历史记录

🛡️ 四层安全防御体系

  • 生成时最小权限
  • 验证时沙箱隔离
  • 入库时版本审计
  • 使用时 Token 预算

🔗 技能谱系全链路追踪

  • 每个技能可追溯来源对话、锻造轮次、验证记录
  • 版本化存储，支持一键回滚
  • 依赖图谱和相关技能推荐

🎯 意图感知智能召回

  • 先分析任务意图，再召回相关技能
  • 四维加权评分 + Token 预算管理
  • 奖励驱动：用得好的技能排前面

📊 增量式经验积累

  • 每轮对话都在积累经验痕迹
  • 经验密度达到阈值自动触发锻造
  • Idle 触发作为兜底机制

🎨 原生 DSH Web UI 集成

  • 右侧挤压式边栏（文件浏览器 / Forge / 统计）
  • 全局顶栏注入
  • Toast 通知 + 对话内审批
  • 统计仪表盘（7 天趋势 / Top 5 / 注入统计）

🛠 技术栈

  • TypeScript strict 模式（0 错误）
  • Cordis 插件架构
  • Zod / Schemastery 结构化校验
  • React 18 + CSS Variables
  • 零额外运行时依赖

📦 安装

dsh plugin --profile web add dsh-skill-forge

🙏 致谢

灵感来源：SkillOpt / EvoSkills / Darwin.skill / Taotie.skill / MemoraX Code

Loading