Token导航 LogoToken导航

Agent Skill · 开发

skill-hub-gateway(技能中心网关)

skill-hub-gateway 用于补充开发相关能力,适合在 OpenClaw 中需要让 Agent 承接开发相关任务时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

skill-hub-gateway 是开发类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为接触敏感数据;许可证为 MIT-0,第三方扫描有提醒项,按安装量在开发类的 21,746 个 Skill 里排第 2,663。

总安装 15,496周安装 633GitHub Stars 1下载 5,013核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
2.4.2

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:skill-hub-gateway(技能中心网关)
来源仓库:https://github.com/chinasilva/skill-hub-gateway
安装命令:
openclaw skills install skill-hub-gateway
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install skill-hub-gateway

这个 Skill 是做什么的

开发类 Skill 面向写代码、改代码和排查问题,一般会读写仓库文件,部分还会执行构建或测试命令。

skill-hub-gateway 作为统一网关,支持异步任务执行、轮询与遥测反馈流程。适用于需要长时间运行或后台处理的开发任务。可集成门户关闭机制,提升用户体验与系统可控性。安装命令为 openclaw skills install skill-hub-gateway。

使用前需配置正确的回调地址与超时参数,避免任务卡死。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install skill-hub-gateway;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

skill-hub-gateway 怎么安装?

在终端执行 openclaw skills install skill-hub-gateway。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

skill-hub-gateway 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

skill-hub-gateway 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

skill-hub-gateway 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

skill-hub-gateway 用的人多吗?

公开数据是累计安装 15,496 次,近一周安装 633 次,下载 5,013 次,GitHub 1 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

skill-hub-gateway 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理开发类任务、又不想每次都手写提示词的人。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 需要厂商级技术支持和 SLA 的生产系统。

开发类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
skill-vetter

技能审查员 · OpenClaw

8,779,018986操作浏览器MIT-0通过
ontology

本体论 · OpenClaw

6,636,960576只读MIT-0通过
entra-app-registration

内部应用程序注册 · Codex、Claude

6,631,872764接触敏感数据未明确通过
entra-app-registration

内部应用程序注册 · Codex、Claude

3,757,680201接触敏感数据未明确通过
nano-banana-pro

纳米香蕉专业版 · OpenClaw

3,526,898364接触敏感数据MIT-0有提醒
skill-hub-gateway

技能中心网关 · 本页

15,4961接触敏感数据MIT-0有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
skill-hub-gateway
description
Unified gateway skill for async execute/poll, portal user closure, and telemetry feedback workflows.
version
2.4.2
metadata
openclaw
skillKey
skill-hub-gateway
emoji
🧩
homepage
https://gateway.binaryworks.app
requires
bins

Skill Hub Gateway

Default API base URL: https://gateway-api.binaryworks.app

Default site base URL (for upload route): https://gateway.binaryworks.app

Chinese documentation: SKILL.zh-CN.md

Version Check Protocol (Agent)

  • Official latest version source: GET /skills/manifest.json -> data.version.
  • Local current version source: this installed SKILL.md frontmatter version.
  • Compare versions using semantic version order (major.minor.patch).
  • Check timing: once at session start, then at most once every 24 hours within the same session.
  • If version check fails (network/timeout/parse error), do not block runtime execution. Continue current workflow and retry at the next allowed check window.

Agent Decision Flow

  • If latest_version > current_version, read the matching section under Release Notes in this document to build update_summary.
  • Agent should show the user:

- current_version - latest_version - update_summary

  • User decision options:

- Update now - Remind me later in this session

  • If user chooses Remind me later in this session, suppress repeated prompts for the same target version until a new session starts.

First-Time Onboarding (install_code)

Scripts auto-complete onboarding by default:

  1. POST /agent/install-code/issue with {"channel":"local"} or {"channel":"clawhub"}.
  2. Read data.install_code.
  3. POST /agent/bootstrap with {"agent_uid":"<agent_uid>","install_code":"<install_code>"}.
  4. Read data.api_key, then call runtime APIs with X-API-Key or Authorization: Bearer <api_key>.

Manual override:

  • You can still provide api_key explicitly.
  • If agent_uid and owner_uid_hint are omitted, scripts derive stable local defaults from the current workspace path.

Legacy Runtime Contract (Compatibility)

  • Execute: POST /skill/execute
  • Poll: GET /skill/runs/:run_id

Portal Actions (User Closure)

Action catalog (single default path per action):

  • portal.me -> GET /user/me
  • portal.balance -> GET /user/points/balance
  • portal.ledger.query -> GET /user/points/ledger
  • portal.usage.query -> GET /user/usage
  • portal.skill.execute -> POST /user/skills/execute
  • portal.skill.poll -> GET /user/skills/runs/:runId
  • portal.skill.presentation -> GET /user/skills/runs/:runId/presentation
  • portal.voucher.redeem -> POST /user/vouchers/redeem (write)
  • portal.recharge.create -> POST /user/recharge/orders (write)
  • portal.recharge.get -> GET /user/recharge/orders/:orderId

Write safety gate:

  • portal.voucher.redeem and portal.recharge.create require payload.confirm === true.
  • If confirm is missing or not true, action runner rejects the call locally and does not send write traffic.

Payload Contract

Default payload conventions:

  • payload.input is the primary input object for portal.skill.execute.
  • payload.request_id is optional and passed through as-is.
  • Query actions use payload fields directly as query params (date_from, date_to, service_id, channel).

Media normalization conventions:

  • Prefer explicit URL fields already present in input: image_url, audio_url, file_url.
  • If attachment.url exists, map it to the capability target URL field.
  • If file_path exists, auto-upload through {site_base}/api/blob/upload, with fallback to {site_base}/api/blob/upload-file when @vercel/blob/client is unavailable, then backfill URL into input before execution.
  • For agent runtimes that do not include @vercel/blob/client, you can also pre-upload media via your backend (for example Vercel Blob) and pass attachment.url or explicit URL fields.
  • site_base_url is a guarded field: runtime only accepts the trusted configured site base URL (default https://gateway.binaryworks.app or env SKILL_SITE_BASE_URL).
  • Users should not need to manually paste media URLs in normal product flows.

Presentation files:

  • portal.skill.presentation accepts optional include_files=true to return visual.files.assets with rendered file URLs.
  • portal-action.mjs defaults include_files=true for portal.skill.presentation unless explicitly disabled.
  • Image runs return overlay (boxed image), plus mask / cutout assets when available.
  • Audio runs normalize data URLs to output.media_files.assets with uploaded file URLs when blob storage is configured.

Runtime Auth Bridge

User-scoped actions use a fixed auth bridge:

  1. Resolve runtime API key context (api_key, agent_uid, owner_uid_hint, base_url).
  2. GET /agent/me with X-API-Key + x-agent-uid to resolve user_id.
  3. POST /user/api-key-login with user_id + api_key to obtain userToken.
  4. Execute Portal Action with Authorization: Bearer <userToken>.

Insufficient Points Recovery

For POINTS_INSUFFICIENT responses:

  • Keep error.code and error.message unchanged.
  • Preserve and surface error.details.recharge_url when available.
  • Diagnostic logs should recommend portal.recharge.create or direct open of recharge_url.

Bundled Files

  • scripts/execute.mjs (CLI args: [api_key] [capability] [input_payload] [base_url] [agent_uid] [owner_uid_hint])
  • scripts/poll.mjs (CLI args: [api_key] <run_id> [base_url] [agent_uid] [owner_uid_hint])
  • scripts/feedback.mjs (CLI args: [api_key] [payload_json] [base_url] [agent_uid] [owner_uid_hint])
  • scripts/telemetry.mjs (shared best-effort telemetry helper)
  • scripts/runtime-auth.mjs (shared auto-bootstrap helper)
  • scripts/portal-auth.mjs (api-key -> user session bridge)
  • scripts/portal-action.mjs (CLI args: [api_key] <action> <payload_json> [base_url] [agent_uid] [owner_uid_hint])
  • scripts/attachment-normalize.mjs (attachment URL/path normalization + upload)
  • references/capabilities.json
  • references/openapi.json
  • SKILL.zh-CN.md

Telemetry and Feedback

  • Bundled scripts now support best-effort telemetry emit for auth/execute/poll/feedback flows.
  • Telemetry is non-blocking and does not change runtime exit semantics if telemetry delivery fails.
  • Optional environment variables:

- SKILL_TELEMETRY_ENABLED (true by default) - SKILL_TELEMETRY_BASE_URL (defaults to runtime base_url) - SKILL_TELEMETRY_TIMEOUT_MS (2000 by default)

  • Feedback script posts to POST /feedback/submit with runtime auth (X-API-Key) and attaches agent_uid + owner_uid_hint in metadata.

Capability IDs

  • human_detect
  • image_tagging
  • tts_report
  • embeddings
  • reranker
  • asr
  • tts_low_cost
  • markdown_convert
  • face-detect
  • person-detect
  • hand-detect
  • body-keypoints-2d
  • body-contour-63pt
  • face-keypoints-106pt
  • head-pose
  • face-feature-classification
  • face-action-classification
  • face-image-quality
  • face-emotion-recognition
  • face-physical-attributes
  • face-social-attributes
  • political-figure-recognition
  • designated-person-recognition
  • exhibit-image-recognition
  • person-instance-segmentation
  • person-semantic-segmentation
  • concert-cutout
  • full-body-matting
  • head-matting
  • product-cutout

Release Notes

When publishing a new version, add a new section here. Agent update summaries must be generated from this block.

2.4.2 (2026-03-12)

What's New

  • Added /api/blob/upload-file direct-upload fallback for file_path flows when @vercel/blob/client is unavailable.
  • Added portal.skill.presentation file rendering (visual.files.assets) for overlay/mask/cutout URLs.
  • Added audio output normalization to return output.media_files.assets with uploaded file URLs when blob storage is configured.

Breaking/Behavior Changes

  • None.

Migration Notes

  • For agent runtimes without @vercel/blob/client, send file_path as before; the runtime now falls back to direct upload.
  • To receive rendered files, call portal.skill.presentation with include_files=true or rely on the portal default.

2.4.1 (2026-03-12)

What's New

  • Added server-side portal.skill.execute input normalization for attachment.url (both input.attachment.url and top-level attachment.url).
  • Added explicit media URL precedence guard (image_url/audio_url/file_url take priority over attachment.url when both are present).
  • Added end-to-end tests for upload-first execution payload compatibility and precedence behavior.
  • Clarified fallback guidance for runtimes missing @vercel/blob/client to use backend pre-upload URLs.

Breaking/Behavior Changes

  • None.

Migration Notes

  • Existing portal-action.mjs workflows remain compatible.
  • For local file workflows in constrained agent runtimes, prefer backend pre-upload then pass attachment.url or explicit media URL.

2.4.0 (2026-03-12)

What's New

  • Added Portal Actions contract for single-skill user closure flows (portal.me, balance/ledger/usage, execute/poll/presentation, voucher and recharge actions).
  • Added local double-confirm write gate: portal.voucher.redeem and portal.recharge.create require confirm=true.
  • Added runtime auth bridge script (portal-auth.mjs) for api_key -> user session.
  • Added action runner (portal-action.mjs) and attachment normalization script (attachment-normalize.mjs) with attachment.url and file_path support.
  • Standardized insufficient-points diagnostics to preserve recharge_url in passthrough envelopes.

Breaking/Behavior Changes

  • Write actions now fail fast locally when confirm=true is absent.

Migration Notes

  • Existing execute.mjs and poll.mjs workflows remain available.
  • Portal users should call portal-action.mjs for user closure actions.

2.3.4 (2026-03-11)

What's New

  • Added best-effort telemetry emit in runtime scripts for auth/execute/poll flows.
  • Added scripts/feedback.mjs to submit structured feedback via runtime auth (X-API-Key).
  • Added shared telemetry helper with optional env controls (SKILL_TELEMETRY_*).

Breaking/Behavior Changes

  • None.

Migration Notes

  • Existing execute/poll calls remain unchanged.
  • If you run scripts in constrained environments, allow optional telemetry calls to /agent/telemetry/ingest and /telemetry/ingest.

2.3.3 (2026-03-11)

What's New

  • Standardized user-facing guidance around upload-based flows: do not expose manual URL/JSON fields for media/document inputs.
  • Updated capability reference examples and wording to match upload-first product UX.
  • Renamed CLI argument label from input_json to input_payload for consistency with structured payload semantics.

Breaking/Behavior Changes

  • None.

Migration Notes

  • Existing runtime API calls remain unchanged.
  • If your integration shows custom input forms, prefer file/attachment inputs instead of manual URL/JSON text fields.

平台分布

OpenClaw

83.8%
按下载量换算4,201

安全审计

VirusTotal

通过

ClawScan

可疑

Static analysis

可疑

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills