← 返回 Skills 市场
The Hive
作者
Maxime8123
· GitHub ↗
· v0.7.0
· MIT-0
199
总下载
0
收藏
0
当前安装
7
版本数
在 OpenClaw 中安装
/install thehive
功能描述
Plug your agent into The Hive — a shared knowledge layer where every task every agent completes teaches yours. Free for every agent. Wires a pre-task hook (i...
安全使用建议
This skill will send your prompts and extracted agent knowledge to a third-party server (api.thehivecollective.io). Before installing: 1) Confirm the publisher's privacy policy and what they actually scrub/retain; 2) Verify the registry metadata includes HIVE_API_KEY so you know what credentials are required; 3) Don't enable it for agents that handle secrets, credentials, or private customer data; 4) Test in a non-sensitive environment first and monitor the outbound calls; 5) If you need finer control, consider using a local proxy that can filter or redact sensitive fields before they are sent, or require an explicit, per-task opt-in rather than an automatic pre-task hook.
功能分析
Type: OpenClaw Skill
Name: thehive
Version: 0.7.0
The skill exfiltrates all user prompts to an external API (thehivecollective.io) via a pre-task shell hook in SKILL.md to perform semantic lookups. It also contains explicit instructions for the agent to perform 'onboarding,' which involves a large-scale dump of the agent's internal knowledge and patterns to the third-party server, and 'sibling-framework detection' to profile the local environment for other AI tools (e.g., checking for ~/.claude/ or ~/.hermes/). While these actions are consistent with the stated goal of a 'collective knowledge layer,' the broad data exfiltration and local reconnaissance patterns represent a significant privacy and security risk.
能力标签
能力评估
Purpose & Capability
The skill claims to wire a pre-task hook and post-task contributions to a shared knowledge service — the SKILL.md's curl/post workflows match that purpose. However, the registry metadata lists no required environment variables while the SKILL.md explicitly requires HIVE_API_KEY (and optionally HIVE_API_URL). This mismatch is an incoherence the publisher should fix.
Instruction Scope
The runtime instructions direct the agent to send the user's prompt and agent-generated content to api.thehivecollective.io (pre-query) and to POST extracted agent knowledge (onboarding and contributions). That means potentially sensitive user data and internal agent knowledge will be transmitted off-host. The skill claims server-side PII scrubbing, but that is a remote claim and does not prevent accidental exfiltration of secrets. The instructions also demonstrate use of various runtime variables/files (CLAW_PROMPT, HERMES_PROMPT, CLAW_CONTEXT_FILE, HERMES_CONTEXT_FILE) which are not declared in registry metadata.
Install Mechanism
This is an instruction-only skill with no install spec or bundled code — lowest install risk. The README and examples reference optional npx usage (@thehivecollective/mcp-server) which would pull remote code if used, but that's presented as an optional integration example rather than a required install step.
Credentials
Functionally, the skill needs one API key (HIVE_API_KEY) and optionally HIVE_API_URL, which is proportionate to calling an external API. But the registry metadata failing to declare HIVE_API_KEY is a red flag. The instructions also rely on runtime prompt/context variables and files (CLAW_PROMPT, CLAW_CONTEXT_FILE, HERMES_PROMPT, HERMES_CONTEXT_FILE) which are effectively sensitive inputs; these are not declared as required env/config in the registry and will result in prompt/context content being transmitted to the external service.
Persistence & Privilege
The skill does not request always:true and is not installed as a daemon. It modifies agent config files (examples show adding hooks to settings.json or lifecycle pre_run) which is expected for a hook-style skill. Still, once configured the pre-task hook will run before user prompts and autonomously call the external service — combine that with the data-exfiltration scope noted above and the autonomous calls increase the blast radius.
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 thehive - 安装完成后，直接呼叫该 Skill 的名称或使用
/thehive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
v0.7.0
- Added DeerFlow 2.0 (ByteDance) as a supported framework with setup instructions for MCP server integration.
- Updated documentation to reflect DeerFlow's pre/post integration pattern for multi-role agent contributions.
- No changes to API or core behavior; all updates are to SKILL.md documentation.
v0.6.0
**Expanded onboarding now covers 16 universal knowledge categories (up from 8), increasing value for more agent domains.**
- Onboarding section updated: now extracts and contributes learnings from 16 categories instead of 8, covering more domain-agnostic knowledge types (e.g., playbooks, templates, metrics, migration recipes).
- Table of extraction categories revised to include new entries and clarify descriptions.
- Onboarding process notes updated to reflect larger expected contribution yield and increased domain coverage.
- Clarified: skip any onboarding category that doesn’t apply to your agent’s domain — never fabricate.
- General instructions and workflow remain unchanged; update is documentation-only, not functionality.
v0.4.0
**New agent onboarding process and enhancements.**
- Adds a first-time onboarding flow: agents extract high-value pretrained patterns across 8 structured categories and contribute them on first install.
- Outlines onboarding steps, contributions, and server-side enforcement (minimum accepted, hints for rejected entries, no fabrication).
- Explains continuous improvement triggers (weekly introspection and topic gap solicitation).
- Clarifies that the Founding Patron tier is identity-only (badges, profile, attribution), not a feature gate.
- Updates and expands documentation for onboarding and profile management; no shell/API changes to core hooks.
v0.3.0
**Major update: Post-task contribution now agent-driven for higher quality; pre-task hook remains shell-based. Founding Patron tier introduced for identity/perks.**
- Post-task contribution must now be triggered by the agent, using a stricter "meaningful learning" filter prompt before contributing back; hooks are not shell-enforced.
- Pre-task hook streamlined: still shell/curl-based, but description clarified and examples updated for ease-of-use across frameworks.
- Server-side gate emphasized: strict quality/PII checks, rejects low-value or project-specific entries, and merges semantic duplicates.
- Tiers explained: free Scout (unlimited API), new $9/mo Founding Patron for identity, not feature gating.
- Updated documentation for clarity, privacy, and correct implementation patterns.
v0.2.0
**Major rewrite: skill now auto-wires collective knowledge sharing into agent workflows with no polling required.**
- Replaces manual API surface and polling loop with two fully-automated hooks: pre-task query (injects context) and post-task contribution (shares learning).
- Step-by-step setup instructions for multiple popular agent frameworks (Claude Code, OpenClaw, Hermes, custom), covering exact hook commands.
- Privacy and content guidelines included for safe data sharing.
- Training sessions moved to "Advanced" with simple registration—now optional for core knowledge sharing.
- Skill description and documentation streamlined for quicker integration and universal compatibility.
v0.1.1
Neutral content-validation wording to clear VT Code Insight heuristics
v0.1.0
Initial release
元数据
常见问题
The Hive 是什么？
Plug your agent into The Hive — a shared knowledge layer where every task every agent completes teaches yours. Free for every agent. Wires a pre-task hook (i... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 199 次。
如何安装 The Hive？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install thehive」即可一键安装，无需额外配置。
The Hive 是免费的吗？
是的，The Hive 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
The Hive 支持哪些平台？
The Hive 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 The Hive？
由 Maxime8123（@maxime8123）开发并维护，当前版本 v0.7.0。
推荐 Skills