lgwanai/openforce

Última atividade de origem registrada: 25 de maio de 2026 às 13:35

Catálogo do SkillsMP atualizado: 12 de novembro de 2026 às 12:52

skills coletadas: 72

Estrelas no GitHub: 4
Forks no GitHub: 1

Instalar com um assistente de IA

Copie este prompt para o assistente de IA que você usa.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/lgwanai/openforce.

README do repositório

Arquivo de origem do README: README.md

Ler o README original

Expandir README completo

# OpenForce [![License](https://img.shields.io/badge/license-OpenForce%20Learning%20v1.0-blue.svg)](LICENSE) [![Rust](https://img.shields.io/badge/rust-1.85+-orange.svg)](https://rust-lang.org) > **道生一，一生二，二生三，三生万物。** — 道德经·第四十二章 **Agent OS** — 蜂群式 AI Agent 编排平台。一个任务进入，Planner 语义理解 → RoundTable 分解 → DAG 调度 → 198 个角色中匹配 N 个 Worker 并行执行。 --- ## 快速开始 ```bash export API_KEY="sk-..." export REDIS_URL="redis://localhost:6379" # 可选 # 普通语义输入 openforce new "分析项目架构并找出安全问题" # 斜杠指令激活 Skill openforce /graphify 分析 src 目录 openforce /code-review openforce /tdd # 多轮 Session openforce approve # 通过 Gate openforce reject "加强认证模块" # 拒绝 + 重规划 openforce continue # 恢复最新 Session openforce skills # 列出 83 个 Skill openforce sessions # 列出所有 Session ``` --- ## 核心架构 ``` 用户输入 (语义 or /skill:name) │ ▼ ┌──────────────────────────────────────────┐ │ Planner — 语义分类 + RoundTable 分解 │ │ ├── Semantic Classification (LLM) │ │ ├── Agent Catalog: 198 角色渐进式披露 │ │ ├── Skill Catalog: 83 Skill Level 1 XML │ │ ├── Round 1: N 维度并行分析 │ │ ├── Round 2: 交叉审查 MECE │ │ └── Round 3: JSON 合成 → TaskTree │ └──────────────┬───────────────────────────┘ │ ▼ ┌──────────────────────────────────────────┐ │ Scheduler — DAG 构建 + Wave 调度 │ │ ├── build_dag(): 依赖推断 + 优先级 │ │ ├── compute_waves(): 拓扑分层 │ │ └── max_concurrent=8: 分片执行 │ └──────────────┬───────────────────────────┘ │ ┌─────────┼─────────┐ ▼ ▼ ▼ ┌─────────┐┌─────────┐┌─────────┐ │Worker-1 ││Worker-2 ││Worker-3 │ │Agent: ││Agent: ││Agent: │ │Backend ││Frontend ││Security │ │Architect││Designer ││Auditor │ │ ││ ││ │ │Native ││Native ││Native │ │Tool Call││Tool Call││Tool Call│ │Loop ││Loop ││Loop │ └────┬────┘└────┬────┘└────┬────┘ │ │ │ └─────────┼─────────┘ ▼ ┌──────────────────────────────────────────┐ │ Redis Session — Todo + Worker Snapshot │ │ session:{id}:todos — 任务进度追踪 │ │ session:{id}:worker:{wid} — Worker 状态 │ └──────────────────────────────────────────┘ ``` ## 关键能力 | 能力 | 实现 | |------|------| | **Agent 角色库** | 198 角色 (来自 agency-agents)，20 领域，渐进式 XML 披露 | | **Skill 系统** | 83 Skills，Level 1 XML → Level 2 on-demand | | **斜杠指令** | `/graphify` `/code-review` `/tdd` 等，精确+模糊匹配 | | **原生 Tool Calling** | OpenAI Function Calling + Anthropic Tool Use，8 工具 | | **对话历史** | Worker 多轮累积 user→assistant(tool_calls)→tool_results | | **RoundTable 规划** | N 维度分析 → 交叉审查 → JSON 合成 | | **DAG 调度** | 拓扑分层 + 优先级 + max_concurrent=8 | | **Todo 追踪** | Redis 持久化，[DONE:n] 标记，Resume 支持 | | **上下文管理** | AgentMemory + Conversation 双重溢出检测 + 结构化压缩 | | **渐进式披露** | Agent: name+desc → full profile; Skill: name+desc → body+scripts | ## Agent 角色 (198) | 领域 | 数量 | 示例 | |------|------|------| | engineering | 29 | Backend Architect, AI Engineer, Code Reviewer, DevOps Automator | | marketing | 30 | Content Creator, SEO Specialist, Bilibili Content Strategist | | specialized | 41 | Blockchain Security Auditor, Compliance Auditor, Customer Service | | design | 8 | UI Designer, UX Architect, Brand Guardian | | testing | 8 | API Tester, Accessibility Auditor, Performance Benchmarker | | sales | 8 | Account Strategist, Sales Engineer, Pipeline Analyst | | paid-media | 7 | PPC Strategist, Programmatic Buyer, Creative Strategist | | support | 6 | Legal Compliance Checker, Support Responder | | spatial-computing | 6 | visionOS Spatial Engineer, XR Immersive Developer | | project-management | 6 | Project Shepherd, Experiment Tracker | | product | 5 | Product Manager, Behavioral Nudge Engineer | | finance | 5 | Financial Analyst, Investment Researcher | | game-development | 5 | Game Designer, Technical Artist, Narrative Designer | | academic | 5 | Anthropologist, Historian, Psychologist | ## Worker Tool Calling 每个 Worker 使用 LLM 原生 Function Calling，8 个工具： | 工具 | 用途 | |------|------| | `read_file` | 读取文件 (16K 截断) | | `write_file` | 写入文件 | | `shell_exec` | 执行 Shell (30s 超时，危险命令拦截) | | `mark_all_done` | 标记所有子任务完成 | | `record_finding` | 记录关键发现 | | `skill_load` | 渐进式加载 Skill body | | `skill_load_ref` | 加载 Skill 参考文件 | | `skill_exec_script` | 执行 Skill 脚本 | Worker 采用**验收标准驱动**的逐子任务执行：每个 subtask 有明确的 acceptance criterion，Worker 执行完自检，达标 (VERIFIED) 进入下一个，不达标 (FAILED) 最多 8 次重试。criteria met = done，不是死循环。 ## Todo 追踪 ``` session:{id}:todos → Redis JSON: items[] → {id, content, status, priority, agent, started_at, completed_at, output_ref} [DONE:n] 标记: LLM 输出 "[DONE:3]" → 自动标记 todo#3 completed Resume: openforce continue → 加载 Redis Todo → 跳过已完成 ``` ## 项目结构 — 24 Crate / 137 源文件 ``` ┌──────────────────────────────────────────────────────────┐ │ 【AI 引擎】 │ │ openforce-cli/ CLI + Planner + AgentRegistry + DAG │ │ llm-client/ OpenAI + Anthropic Function Calling │ │ knowledge-base/ 语义分类 (ExpertIndex) │ │ skill/ SKILL.md: discovery→parser→executor │ ├──────────────────────────────────────────────────────────┤ │ 【控制面】 │ │ domain/ Session/Event(30)/Command(15)/Lease │ │ session-store/ Event Sourcing + CAS + Projection │ │ scheduler/ DAG + LeaseIssuer + CapabilityToken │ │ policy-engine/ 三层授权 (mTLS+Token+业务) │ │ redis-session/ Redis Session + TodoList + Snapshot │ │ proto/ gRPC 服务定义 │ │ gateway/ REST→gRPC 代理 │ ├──────────────────────────────────────────────────────────┤ │ 【数据面】 │ │ worker/ criteria-driven tool calling (583行) │ │ node-daemon/ Worker 生命周期管理 │ │ cube-sandbox/ Firecracker MicroVM (9 文件) │ │ space-manager/ 三层隔离 + WarmPool │ ├──────────────────────────────────────────────────────────┤ │ 【安全基础设施】 │ │ mtls/ Ed25519 CA + SPIFFE + 证书轮换 │ │ path-acl/ 路径 ACL + 规范化防遍历 │ │ patch-classifier/ Patch 风险分级 (9 PCR) │ │ project-tools/ HITL 审批 │ │ effect-gateway/ 副作用网关 (幂等 + Outbox) │ ├──────────────────────────────────────────────────────────┤ │ 【进化面 + 多租户 + 监控】 │ │ evolution/ Observer + Evaluator + Evolver │ │ launch-checker/ 上线验证 (ReleaseGate + RedTeam) │ │ tenant-governance/ BYOK + 配额 + 留存 + 离场 │ │ tui-dashboard/ Ratatui 终端面板 │ └──────────────────────────────────────────────────────────┘ ``` ## 技术栈 | 组件 | 技术 | |------|------| | 语言 | Rust 2024 | | LLM | OpenAI / Anthropic Native Function Calling | | 存储 | Redis (Session + Todo) | | 传输 | gRPC (tonic) + HTTP (reqwest) |

Carregando o README do repositório…

Skills neste repositório

Desenvolvedores de software Analistas de garantia de qualidade de software e testadores Designers de interfaces web e digitais Outras ocupações de informática

6 categorias ocupacionais · 99% classificado

Desenvolvedores de software

CRUD for phases in ROADMAP.md — add, insert, remove, or edit phases

Idioma do texto original: inglês

2026-05-24

Mostrando 40 de 72 skills coletadas.