← 返回 Skills 市场
173
总下载
0
收藏
1
当前安装
7
版本数
在 OpenClaw 中安装
/install verified-humanizer
功能描述
Transform AI-generated content into natural, human-sounding writing, measure the improvement, and optionally verify the result.
安全使用建议
This package appears coherent and privacy-focused: it keeps rewriting and evaluation local and does not request credentials or external services. Before installing, verify that any external verification you choose to integrate (if you enable verification later) is configured to accept only structured metrics (no raw or rewritten text). Also review and sandbox the included hook scripts so the system message they inject matches your agent policy before you enable the skill in production.
功能分析
Type: OpenClaw Skill
Name: verified-humanizer
Version: 0.0.7
The skill bundle is a text-transformation tool designed to improve AI-generated content while prioritizing privacy. All components, including the shell scripts (scripts/scan-text.sh) and OpenClaw hooks (hooks/openclaw/handler.js), are aligned with the stated purpose of local text analysis and explicitly instruct the agent to avoid sending raw data to external systems. No indicators of data exfiltration, malicious execution, or harmful prompt injection were found.
能力评估
Purpose & Capability
Name/description (humanizing AI text, measuring improvements, optional verification) aligns with provided files: rewrite guidance (SKILL.md, README), evaluation templates, local helper scripts, and lightweight OpenClaw hook. There are no unrelated dependencies, credentials, or binaries requested.
Instruction Scope
SKILL.md explicitly requires local rewriting, local evaluation, and that verification (if used) only transmit structured metrics — and the included scripts/hooks follow that guidance. No instructions request reading unrelated system files, secrets, or sending raw text externally.
Install Mechanism
No install spec is provided (instruction-only style). The package includes small local scripts and hook files; nothing downloads or executes external code or archives. This is low-risk and proportionate to the stated purpose.
Credentials
No environment variables, credentials, or config paths are required. The skill does not ask for unrelated secrets and the SKILL.md forbids sending private text to verifiers, so requested access is minimal and appropriate.
Persistence & Privilege
always is false and the hook only returns a brief system reminder (and skips when context.subagent is present). The skill does not request persistent/global privileges or modify other skills' config.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install verified-humanizer - 安装完成后，直接呼叫该 Skill 的名称或使用
/verified-humanizer触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v0.0.7
- Skill documentation rewritten for clarity and brevity
- Simplified step descriptions and execution loop
- Streamlined output format section with example input/output
- Reduced repetition and removed detailed pattern guides
- Verification and privacy notes made more concise
- No changes to code or functionality; documentation update only
v0.0.6
* Converted Verified Humanizer from wrapper-style to a full packaged skill structure.
* Rewrote `SKILL.md` for clarity, scan safety, and better ClawHub presentation.
* Preserved core model: rewrite → evaluate → (optional) verify.
* Integrated optional SettlementWitness-style verification at the evaluation stage using structured metrics only.
* Removed required identity (`agent_id`) and any implication of credentials or external dependency.
* Ensured verification is optional, user-controlled, and limited to non-sensitive data (evaluation metrics only).
* Added packaged skill components:
* `_meta.json`
* `.learnings/` (errors, feature requests, learnings)
* `assets/` (templates for consistent outputs)
* `hooks/` (OpenClaw bootstrap guidance)
* `references/` (usage and integration notes)
* `scripts/` (lightweight helpers, scan-safe)
* Improved evaluation clarity with structured before/after metrics and delta reporting.
* Optimized for ClawHub scanner:
* no explicit external API calls
* no required identity
* no credential handling
* Improved README for clearer positioning and adoption.
* No breaking changes to core functionality; improvements focus on structure, safety, and usability.
v0.0.5
Improved summary and top-level description for clarity and higher conversion. No functional changes.
v0.0.4
Restricted attestation payload to evaluation.delta only and explicitly prohibited transmission of original or rewritten text to prevent data leakage.
v0.0.3
Added SettlementWitness endpoint and operator details, clarified attestation data flow, and resolved external verification transparency requirements.
v0.0.2
Clarified attestation flow, defined settlement_witness behavior, and removed ambiguity around agent identity and credential usage.
v0.0.1
Introduced measurable before/after evaluation for text transformation with optional SAR attestation and improved trust-safe positioning.
元数据
常见问题
Verified Humanizer 是什么？
Transform AI-generated content into natural, human-sounding writing, measure the improvement, and optionally verify the result. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 173 次。
如何安装 Verified Humanizer？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install verified-humanizer」即可一键安装，无需额外配置。
Verified Humanizer 是免费的吗？
是的，Verified Humanizer 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Verified Humanizer 支持哪些平台？
Verified Humanizer 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 Verified Humanizer？
由 nutstrut（@nutstrut）开发并维护，当前版本 v0.0.7。
推荐 Skills