功能描述

ASF V4.0 工业化增强模块 - 治理门禁 + 成本模型 + 安全优化。提供否决权执行、所有权证明、经济学评分、返工风险预测、安全在线优化等工业级能力。

使用说明 (SKILL.md)

ANFSF V1.5.6 Skill - Enhanced Hybrid Adaptive Parser

Name: Asf V4
Author: alshowse-tech

AI Native Full-Stack Software Factory V1.5.6 - Enhanced Hybrid Adaptive Parser with 100% health status (187/187 tests)

Overview

This skill provides enhanced requirement refinement capabilities for ANFSF (AI Native Full-Stack Software Factory) architecture, featuring:

Hybrid Adaptive Parser: Smart detection of complex requirements with weighted scoring system
Negation Handling: Avoids false positives (e.g., "不需要多级审批" is not detected as complex)
Multi-format Support: Supports Mermaid, PlantUML, and image references
Template Matching: Automatically matches historical templates
Complete Rollback: Automatic rollback mechanism with monitoring

Key Features

1. Hybrid Adaptive Parser

Intelligently detects complex requirements using weighted scoring system:

Multi-level审批 workflow: +3 points
Cross-department collaboration: +1 point per department
Complex data visualization: +2 points
External system integration: +2 points
Complex permission system: +2 points

2. Negation Processing

Detects negative keywords to reduce false positives:

"不需要" (not required)
"无需" (no need)
"不要" (don't want)
"no", "without", "not required"

3. Multi-format Content Support

Mermaid flowcharts
PlantUML diagrams
Image references (OCR preparation)
Markdown tables and lists

4. Template Matching

Automatically matches historical templates:

fixed-asset-investment (固定资产投资)
project-management (项目管理)
hr-system (人力资源系统)

Usage

The skill automatically selects the appropriate parsing strategy based on requirement complexity:

const skill = new RequirementRefinerSkill(context);
const result = await skill.refine(rawRequirement);

Architecture

Layer 8.5 Governance: Full governance control plane
MemPalace Memory: Enhanced memory architecture with isolated wings
Evolution Harness: Supported for self-calibration

Installation

clawhub install @anfsf-v1/[email protected]

Requirements

Node.js >= 20.0.0
OpenClaw >= 2026.3.24

License

MIT

安全使用建议

What to consider before installing: - Inconsistency flag: the registry says no env vars required, but shipped scripts mention ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY and print instructions to set them. Do not assume 'none' — inspect files for hidden requirements. - Review before running scripts: scripts/update-providers-config.js reads/writes ../../../openclaw.json (it modifies platform provider/fallback configuration). Do not run that script unless you understand and approve the exact changes and have backups of your openclaw.json. - Audit network endpoints: the package references external providers (openai, anthropic, deepseek). Confirm endpoints are legitimate and that keys are managed securely. - Inspect code for config writes: search repository for filesystem writes and network calls (especially any code that mutates OpenClaw configs or memory). Pay attention to any file paths outside the skill directory. - Sandbox first: install and run in an isolated environment or staging instance. Verify behavior (no unexpected provider additions, no outbound calls) before deploying to production. - If you lack capacity to audit: ask a developer or security reviewer to check the following files at minimum: scripts/update-providers-config.js, config/asf-v4.config.yaml, integrations/* (memory-extension.ts, agent-status-extension.ts), and any code that references openclaw.json or other system-level paths. Why suspicious (short): declared metadata understates required credentials and side-effects; shipped code can modify global OpenClaw configuration (provider list/fallbacks) and write to platform memory/status — operations that should be explicit and limited. Proceed only after code review or in a sandbox.

功能分析

Type: OpenClaw Skill Name: asf-v4 Version: 1.5.6 The skill bundle implements a comprehensive 'AI Native Full-Stack Software Factory' (ANFSF) framework focused on governance, security, and requirement refinement. It includes robust security features such as a constitution validator (src/constitution-validator.ts) to enforce safety boundaries and a security sandbox (src/security-sandbox.ts) for isolated code execution. While the bundle contains scripts that modify configuration files (scripts/update-providers-config.js) and perform network probes for style diagnostics (tools/style-diagnostic.ts), these actions are transparently documented, well-tested, and directly support the stated purpose of the skill. No evidence of intentional malice or unauthorized data exfiltration was found.

能力标签

cryptocan-make-purchasesrequires-sensitive-credentials

能力评估

ℹ Purpose & Capability

The skill's name, SKILL.md, and skill.yaml describe governance/veto/ownership/economics features and the repository contains many corresponding modules (veto enforcer, ownership proof, memory and agent-status integrations). That is broadly coherent. However the manifest/registry metadata claims 'required env vars: none' and 'no required binaries' while included scripts and config mention provider API keys and write to OpenClaw config files — a mismatch between declared requirements and actual code.

⚠ Instruction Scope

SKILL.md provides usage and high-level architecture only, but the shipped code and scripts include operations that touch host configuration (scripts/update-providers-config.js reads/writes ../../../openclaw.json), add provider entries, and enable writing to platform memory/status (config/asf-v4.config.yaml). The runtime instructions in SKILL.md do not call out these system config modifications or the environment variables the scripts expect, granting the skill broad discretion that isn't documented in the user-facing instructions.

ℹ Install Mechanism

No install spec is declared (the package is effectively distributed as code in the bundle), which avoids remote downloads. That's lower install risk than a remote fetch, but it's inconsistent with SKILL.md's 'clawhub install' hint. The package contains many files and scripts that will be present on disk after install—so review before running any included scripts.

⚠ Credentials

Registry metadata says 'Required env vars: none' and 'Primary credential: none', yet scripts (e.g., scripts/update-providers-config.js) reference ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY and print warnings about them. The skill's code/config also declares integrations that write to OpenClaw memory/status and may accept provider configuration. This mismatch (no declared env vs code expecting provider keys and being able to modify agent/provider config) is disproportionate and suspicious.

⚠ Persistence & Privilege

always:false (good), but the skill includes integrations/config that will write change events to OpenClaw memory and extend agent status (integration flags in asf-v4.config.yaml). More concerning: an included script targets ../../../openclaw.json (system config) to add/modify model providers and fallback chains, which can alter which model endpoints the platform uses. That level of system/config modification is powerful and should be clearly declared and reviewed before use.

版本历史

v1.5.6

v1.5.6 - 100% health status (187/187 tests)

v1.5.5

Major upgrade: Enhanced Hybrid Adaptive Parser with advanced requirement analysis and multi-format support. - Introduces a Hybrid Adaptive Parser for accurately detecting complex requirements with a weighted scoring system. - Adds negation handling to prevent false positives in requirement detection. - Supports parsing of Mermaid, PlantUML, image references, and Markdown content. - Enables automatic template matching for common business scenarios. - Implements full rollback mechanism and monitoring for refinement operations. - Upgrades architectural documentation and adds comprehensive integration and optimization reports.

v1.5.4

feat: [email protected] - BailianClient v2.0.0 优化版路径二 + 智能兜底新增功能: - qwen3-asr-flash-filetrans 模型 (支持 file_urls) - URL 有效性预检 (HEAD 请求) - 观点提炼 (_extract_points) - 兜底路径 (_fallback_to_path_one) - transcribe_streamlined() 一站式流程优化效果: - 主路径延迟: 8-12s → 3-5s (-60%) - 成功率: 85% → 95%+ - OSS 流量: 100% → 5% (-95%)

v1.5.3

feat: [email protected] - MemPalace 匹配度 100/100 ✅\n\n新增示例:\n- 示例 7: 高级混合检索 (语义+时间+层级)\n- 示例 8: 实体时间线查询 \n- 示例 9: 批量导入和统计\n- 示例 10: 捷阅证券项目完整用例\n\nMemPalace 匹配度: 100/100 (10/10 完整指南)

v1.5.2

docs: v1.5.2 - README update for MemPalace integration - MemPalace architecture documentation added - Hierarchical retrieval examples added - Package metadata updated with MemPalace tags - README: 7.3KB → 8.1KB

v1.5.1

feat: MemPalace 架构融合 - 完整实现 - Wings + Rooms + Halls + Tunnels 结构 - TemporalKG (时间索引) - 层级检索器 - 嵌入器适配器 (OpenAI/本地) - MemPalace 匹配度: 95/100 - 检索精度提升: +25%

元数据

Slug asf-v4

版本 1.5.6

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 6

常见问题

Asf V4 是什么？

ASF V4.0 工业化增强模块 - 治理门禁 + 成本模型 + 安全优化。提供否决权执行、所有权证明、经济学评分、返工风险预测、安全在线优化等工业级能力。它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 150 次。

如何安装 Asf V4？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install asf-v4」即可一键安装，无需额外配置。

Asf V4 是免费的吗？

是的，Asf V4 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Asf V4 支持哪些平台？

Asf V4 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Asf V4？

由 alshowse-tech（@alshowse-tech）开发并维护，当前版本 v1.5.6。

Asf V4