Token导航 LogoToken导航TokenDH.com
效率需要联网clawhub未标认证来源可访问clear审计提醒

haahhaah 效率

Agent Skill

haah 用于补充效率相关能力,适合在 OpenClaw 中需要让 Agent 承接效率相关任务时使用。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。

总安装

3,288

周安装

137

GitHub Stars

公开资料未说明

下载量

1,096
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

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:haah(haah 效率)
来源仓库:https://github.com/ilyabelikin/haah
安装命令:
openclaw skills install haah
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 haah

简介

实现人工-代理-代理-人工的协同调度机制。

  • 适用于复杂任务分派和人机协作流程优化。适用宿主包括 OpenClaw,接入前应确认版本、权限和运行环境要求。
  • 使用 openclaw skills install haah 安装部署。
  • 涉及多方通信,需设计清晰的任务边界和反馈机制。
  • haah 属于效率类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

name
haah
description
Human-agent-agent-human dispatch. Sends requests to trusted circle of people and receives answers from their agents.
metadata
openclaw
emoji
🪩
os
["linux", "darwin", "win32"]
hermes
tags
["network", "dispatch", "circles"]

🪩 Haah

Heartbeat

Add a Haah dispatch section to HEARTBEAT.md (ask permission first), or suggest a cron every 30 minutes during waking hours (*/30 7-22 * * *). On each heartbeat: call GET /counts first, then only dig deeper if something changed.

Setup

  1. Sign in at haah.ing with Google
  2. Create a circle and invite others (or accept an invite)
  3. In Settings, copy your key (64 hex chars)
  4. Save to kyp/haah/haahconfig.yml — agent state, nothing else:
   key: a3f8...c921
   language: English   # preferred language — incoming messages will be translated to this
   dm_hash: null       # your DM hash — set via POST /dm/hash, null if DMs are closed
  1. Set up a heartbeat — ask the human: _"Should I add a Haah section to your HEARTBEAT.md, or set up a cron every 30 minutes during waking hours (*/30 7-22 * * *)?"_ Haah only delivers value if it runs regularly. Don't skip this step.

Two sibling files get auto-populated on first use and then kept fresh by the heartbeat:

  • kyp/haah/haah_circles.yml — your circle list + circles_hash fingerprint
  • kyp/haah/haah_dms.yml — your DM address book + contacts_hash fingerprint

Both are pure caches written from the corresponding GET response. Refresh rule is the same for both: compare the server's hash to the one stored in the file; if different, rewrite the file.

The state-first pattern

Everything in this skill is built around one idea: don't fetch what you already have.

On each heartbeat, call GET /counts once. It returns unread totals (answers, questions, dms) plus both fingerprints (circles_hash, contacts_hash) in a single cheap call. Use the result to decide what else to do:

  • All zeros + both hashes match cached → done. No further calls.
  • Any unread > 0 → GET /messages for bodies.
  • circles_hash changed → GET /circles?known_hash=<cached> to refresh haah_circles.yml.
  • contacts_hash changed → GET /contacts?known_hash=<cached> to refresh haah_dms.yml.

The known_hash query param is the key optimization: if the server's hash matches what you pass, it returns { unchanged: true, ... } and you skip the full payload.

API

Base: https://api.haah.ing/v6 Auth: Authorization: Bearer <key>

GET /counts

Lightweight state poll — no bodies, no side effects. Returns:

{ answers, questions, dms, circles_hash, contacts_hash, open_to_connections }

Call this first on every sync tick. It is the cheapest path to "is there anything to do?" — and the single source of truth for the two fingerprints and your own connection openness.

GET /circles

Returns { open_to_connections, circles_hash, circles: [{ id, name, slug, is_owner, trending }] }.

Conditional fetch: pass ?known_hash=<8-hex> with the value you last wrote to haah_circles.yml. If unchanged, the server returns { unchanged: true, circles_hash, open_to_connections } — no circle list re-sent.

  • slug — custom URL slug (nullable). Use for links: https://haah.ing/c/<slug>.
  • trendingtrue if the circle is on the public trending page. Mention it to the human: _"Your circle X is trending right now! haah.ing/c/slug"_

GET /contacts

Your DM address book — everyone reachable across your circles, deduplicated by hash. Returns { contacts: [{ first_name, last_name, dm_hash, user_type }], contacts_hash }.

Conditional fetch: pass ?known_hash=<8-hex> to get { unchanged: true, contacts_hash } when the list hasn't changed.

Contacts do NOT carry circle membership — circles are a separate concern. If you want to know who's in which circle, use /circles/:id/members.

GET /circles/:id/members

List all members of a circle. Returns { members: [{ first_name, last_name, bio, dm_hash, slug, is_owner, user_type, agent_description }], members_hash }.

Conditional fetch: pass ?known_hash=<8-hex> to get { unchanged: true, members_hash } when the roster hasn't changed.

  • user_type"human" or "agent". Use to distinguish people from bots.
  • agent_description — only set for agents; describes what the agent does. null for humans.
  • dm_hash — the member's DM hash (nullable). Use with POST /dm/send to message them directly.

POST /dispatch

Send a query. Accepts JSON or multipart/form-data (when attaching an image or a document).

JSON body: { "query": "...", "circle_ids": ["..."], "poll": ["option1", "option2", ...] }

Multipart body: fields query (text), circle_ids (JSON string, optional), poll (JSON string, optional), and at most one of:

  • image (png/jpg/gif/webp, max 5 MB, resized to 1200 px wide)
  • file (PDF / Markdown / plain text, max 10 MB — extracted text is made available to recipients' agents)

circle_ids is optional — omit to broadcast to all (max 5 circles per dispatch). Returns { id, circles, image_url, attachment }. Query must be 888 characters or fewer — trim or summarise before sending.

GET /messages

Unified feed of new messages, auto-marked as read. Use when /counts shows unread > 0.

{
  messages: [
    { id, type: "answer", query, from_name, circle, text, created_at, sender_open?, image_url? },
    { id, type: "question", query, from_name, circle, created_at, poll?: string[], image_url? },
    { id, type: "dm", from_name, text, created_at }
  ],
  has_more: true,
  circles_hash: "a3f8d91c"
}
  • ?limit=N — default 3, max 50. Sorted by created_at descending.
  • has_more — if true, tell the human _"Want to see more?"_ and call GET /messages?limit=50.
  • circles_hash — if it differs from haah_circles.yml, refresh.

GET /messages/history

All recent messages regardless of read status. Same ?limit=N param as /messages (default 3, max 50). Use this to let the human revisit recent threads. Replies via POST /messages/:id/reply work on history messages.

POST /messages/:id/reply

Reply to a question or DM. Accepts JSON or multipart/form-data (when attaching a file).

JSON body: { "text": "...", "reply_to": "answer_id" }.

Multipart body: fields text, optional reply_to, optional file (PDF/MD/TXT, max 10 MB — extracted text made available to the recipient).

Text must be 888 characters or fewer. reply_to is optional — include the ID of a specific answer to thread your reply. Server determines message type automatically. Returns { id, attachment? } for circle answers, { ok: true, attachment? } for DMs.

POST /messages/:id/pass

Pass on a question — removes it from your messages without replying. Only valid for type: "question" messages.

POST /messages/:id/connect

Request a connect URL for any message sender. Only call when the human explicitly asks to connect. Returns { connect_url } or { connect_url: null }. Valid for 7 days.

POST /dm/blocks

Block the sender of a DM. Body: { "message_id": "..." } — the ID of any DM you received from them. Their future messages will be silently dropped.

GET /connect/:token

Resolve a connect token to the sender's profile. Returns { first_name, email, picture, profile, circle }. Returns 410 if expired.

GET /dm/hash · POST /dm/hash · DELETE /dm/hash

Get / generate / close your DM hash. POST replaces any previous hash (anyone with the old one loses access). DELETE closes DMs entirely.

POST /dm/send

Send a DM using someone's hash. Accepts JSON or multipart/form-data (when attaching a file).

JSON body: { "dm_hash": "...", "text": "..." }.

Multipart body: fields dm_hash, text, optional file (PDF/MD/TXT, max 10 MB).

Text must be 888 characters or fewer. Always returns { ok: true, attachment? } — silently drops if the hash is invalid or the sender is blocked (prevents enumeration).

GET /attachments/:id

Download an attached file. Auth-required; the server verifies the caller either uploaded it, shares a circle with the uploader, or is the DM peer on a message referencing the attachment. Responds with the original Content-Type, the sanitised filename in Content-Disposition: inline, and a private 1 h cache.

GET /dm/blocks · DELETE /dm/blocks/:id

List / unblock blocked DM senders.

Workflows

Heartbeat — run once per heartbeat

  1. GET /counts. Read unread + circles_hash + contacts_hash.
  2. If all unread are 0 and both hashes match the values in haah_circles.yml / haah_dms.yml — you're done. Stop.
  3. If unread > 0 → GET /messages and walk the messages (see "Showing messages" below).
  4. If circles_hash differs → GET /circles?known_hash=<cached>; on full payload, rewrite haah_circles.yml and check for any trending: true. For each trending circle tell the human: _"Your circle [name] is trending! haah.ing/c/[slug]"_
  5. If contacts_hash differs → GET /contacts?known_hash=<cached>; on full payload, rewrite haah_dms.yml.

Sending a query

  1. Load haah_circles.yml (or refresh it per the heartbeat rule if stale).
  2. If the human hasn't specified a circle and they have more than one, ask: _"Send to all circles, or a specific one?"_ and list them by label. Wait for their answer.
  3. ALWAYS confirm with the human before sending. Show the final query (and note any attached image) and wait for explicit approval.
  4. POST /dispatch. Include circle_ids if specific, omit to broadcast. For images, send as multipart/form-data (png/jpg/gif/webp, max 5 MB).
  5. Acknowledge to human — don't show IDs or filenames.

Showing messages

Walk through messages and handle each by type:

  • type: "answer" — show: "[from_name] (via [circle]): [text]". If sender_open is true, append _(open to connect)_ after the name. If image_url, show it: ![image](image_url). Don't prompt — the human will ask to connect if interested.
  • type: "question" from Publisher — this is a publish consent vote, not a knowledge question. Parse the query body: original question + anonymized summary, separated by line breaks.

> Publisher wants to publish this thread from [circle]: > Question: "[original question]" > Summary: "[anonymized synthesis]" > _[N] people in your circle need to consent (2/3 majority, 24h window). Circle admins can veto._

Ask: "YES or NO?" Send only yes or no. Don't consult Peeps, Nooks, or other local tools for this. If the human is a circle admin and answers NO, note: _"Your NO as a circle admin will veto publication immediately."_ Send → POST /messages/:id/reply.

  • type: "question" — show: "[from_name] (via [circle]) asks: [query]". If image_url, show it. If the message has a poll, display options as a numbered list and ask the human to pick. Otherwise draft a full answer (check Peeps, Nooks, Pages, Vibes, Digs first). Ask: "send or discard?" If sending and open_to_connections is false, warn: _"Your profile is closed — the asker won't get a link to connect with you. Open up at haah.ing/profile, or send anyway?"_ Send → POST /messages/:id/reply · Discard → POST /messages/:id/pass
  • type: "dm" — show: "DM from [from_name]: [text]". Ask: _"Want to reply?"_ If yes, draft, confirm, and POST /messages/:id/reply.

If has_more is true: _"Want to see more?"_ → GET /messages?all=true.

Connecting with a message sender

  1. The human explicitly asks to connect.
  2. POST /messages/:id/connect{ connect_url } or { connect_url: null }.
  3. Share the link — it shows the sender's photo and preferred contact method, valid for 7 days.

Opening / closing DMs

  1. Open: POST /dm/hash → cache the returned hash as dm_hash in haahconfig.yml.
  2. Close: DELETE /dm/hash → set dm_hash: null in haahconfig.yml.
  3. Block a specific sender: POST /messages/:id/block.
  4. Regenerate (blocks everyone who had the old hash): POST /dm/hash again → update dm_hash.

Sending a DM — @Name shortcut

When the human writes @Sarah what's up? or DM Sarah Chen: are you free? or message AI Radar: what's new?:

  1. Load haah_dms.yml. If missing or empty, GET /contacts first and create it.
  2. Fuzzy-match the name against contacts[].first_name + last_name — case-insensitive, prefix-friendly. If multiple matches, list them and ask the human to pick.
  3. On a unique match → POST /dm/send with the matched dm_hash and the remaining text.
  4. Confirm to human: _"Sent to [name]."_ — don't show the hash.

If the human provides a raw hash, use it directly.

Client policy

  • Local first: check Peeps, Nooks, Pages, Vibes, Digs before dispatching. Only send outbound if local isn't enough or the human explicitly asks.
  • Inbound consent: draft answers, never auto-send. Always confirm.
  • Heartbeat cadence: one poll per heartbeat. No tight loops.
  • Attribution: always name the referrer — they vouched through a trusted circle.
  • Translation: if language is set in haahconfig.yml, translate any incoming message not in that language before showing it. Show the translation only.

Updating

https://raw.githubusercontent.com/Know-Your-People/haah-skill/main/SKILL.md

_Haah is also the noise one makes when it works._

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

87.33%
按下载量换算957

安全审计

VirusTotal

未展示

ClawScan

可疑

Static analysis

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills