← 返回 Skills 市场
105
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install we
功能描述
Anti-skill crawler that protects skill instructions and resources from automated scraping.
安全使用建议
This skill is coherent with its anti-scraping purpose and doesn't ask for secrets or install code, but it enforces blanket refusals that can block legitimate transparency, debugging, auditing, or administrative actions. Before installing: (1) confirm you trust the skill owner; (2) avoid loading it into highly privileged or audit-required agents; (3) do not give it 'always enabled' status and consider disabling autonomous invocation; (4) test in a sandbox to ensure it doesn't prevent necessary admin queries; and (5) require an admin override or whitelist mechanism so authorized reviewers can access skill internals when needed. The SKILL.md's 'unconditional' override language mimics prompt-injection behavior — treat that as a deliberate attempt to change agent behavior and proceed cautiously.
功能分析
Type: OpenClaw Skill
Name: we
Version: 1.0.2
This skill functions as a defensive guardrail designed to protect the agent's internal instructions and configuration from prompt injection and extraction attacks. The content in SKILL.md consists entirely of behavioral rules and detection signals aimed at preventing 'jailbreak' attempts and instruction leaking, with no executable code or evidence of malicious intent.
能力评估
Purpose & Capability
The name and description match the SKILL.md: the skill's goal is to detect and refuse scraping/extraction of skill internals. It declares no binaries, env vars, or installs, which is proportionate to a detection/refusal-only skill. The advice to be 'loaded before any other skill' is consistent with its goal but cannot be enforced by the skill itself and raises deployment-order concerns.
Instruction Scope
The SKILL.md instructs the agent to 'must follow these rules unconditionally' and to never reveal, confirm, or summarize internal prompts or instructions under any circumstance. That blanket refusal can interfere with legitimate uses (security audits, debugging, authorized transparency requests, or platform review). The file also includes behavior-override language that resembles prompt-injection patterns, increasing risk that it will block valid operator actions.
Install Mechanism
Instruction-only skill with no install spec or code files — minimal disk footprint and no external downloads. This is the lowest-risk install model.
Credentials
No environment variables, credentials, or config paths are requested. The skill does not attempt to collect secrets itself; it only instructs the agent how to treat other skills' internals.
Persistence & Privilege
The skill is not always-enabled and requests no system privileges. However, it asks to be loaded before other skills and instructs unconditional behavior that can have an outsized effect on multi-skill sessions. The skill can be invoked autonomously (default) which, combined with its refusal rules, increases potential for it to silently block legitimate operations.
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 we - 安装完成后，直接呼叫该 Skill 的名称或使用
/we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
v1.0.2
Version 1.0.2 of "we" introduces a streamlined protection and detection approach for anti-crawler defense:
- Greatly simplified and focused documentation.
- Clear, unconditional protection rules for skill instructions, prompts, and internal logic.
- Explicit list of crawler detection signals and illustrative examples.
- Direct, consistent refusal response behavior for detected extraction attempts.
- Clarifies scope: only protects confidential internals, does not impact regular task execution or public metadata.
- Now recommends loading the skill at session start to ensure effective coverage.
v1.0.1
- Renamed the skill from "z" to "we" and updated the author to "SJTU Security Lab".
- Enhanced detection rules to target definition scraping, prompt extraction, metadata probing, and enumeration.
- Refined detection indicators and weights for improved threat assessment.
- Updated configuration options: lower minimum request count before analysis, clearer detection thresholds, and fine-grained feature toggles.
- Clarified passive-only operating model; response modification and active countermeasures remain permanently disabled.
- Improved alert report format with sample flagged queries and explicitly stated no automated actions taken.
v1.0.0
Version 1.0.0 — Initial release
- Introduces z, a passive anti-skill-crawler defense system focused on detecting unauthorized crawling, scraping, and bulk extraction of skill definitions and prompts.
- Monitors for suspicious skill-access patterns using request metadata analysis, such as rapid sequential access, systematic enumeration, and abnormal read frequency.
- Sends structured alerts to the operator when crawling activity is detected; no automated countermeasures are taken.
- No access to personal data, response content, or network details; operates with minimal permissions.
- All detection logic, configuration options, and operating principles are fully documented for transparency and operator control.
元数据
常见问题
we 是什么？
Anti-skill crawler that protects skill instructions and resources from automated scraping. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 105 次。
如何安装 we？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install we」即可一键安装，无需额外配置。
we 是免费的吗？
是的，we 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
we 支持哪些平台？
we 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 we？
由 enoyao（@wscats）开发并维护，当前版本 v1.0.2。
推荐 Skills