원클릭으로 Manus에서 모든 스킬 실행

holo-qa

스타36

포크7

업데이트2026년 6월 19일 09:00

Use when asked to QA Nimbus Desk, run a qa/*.md behavioral spec, verify UI behavior, check a regression, or confirm a UI fix. Drives the running app through the root Holo MCP tool, holo_desktop(task), and returns a QA report.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

hcompai

hcompai/holo-desktop-cli

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

holo-qa: behavioral QA through root Holo MCP

Use the normal installed Holo MCP tool: holo_desktop(task). Claude Code should be set up with uv run holo install claude-code from the holo-desktop-cli checkout. Do not configure or launch a project-local QA MCP server, and do not use a separate QA runner; this example intentionally has one desktop-agent path.

Workflow

Ensure the Nimbus Desk dev server is running on http://localhost:5173. If it is not, start npm run dev in the background from this directory and wait until curl -s -o /dev/null -w "%{http_code}" http://localhost:5173 returns 200.
Pick the closest spec from qa/*.md. If the user names a spec, use exactly that file. Otherwise map the request by filename and H1 title: chat-* covers the support chat widget, auth-* covers login/session behavior, and tickets-* covers the ticket list.
Read the full markdown spec, including YAML frontmatter. Use the url, viewport, timeout_s, and credentials fields when composing the Holo task.
Call holo_desktop(task) once with a self-contained task using the template below. Wait for the final result.
Treat the Holo report as the UI evidence. Do not manually browse the app as a substitute. Quote failures verbatim.
If asked to fix a failed flow, fix the app code, then rerun the same spec once. If the rerun still fails, stop and summarize both reports.

Holo task template

You are a meticulous read-only black-box QA tester for Nimbus Desk.
Judge only what is visible on screen. Do not open developer tools, read source
code, inspect the DOM, or fix anything. Your job ends at observing and
reporting.

Ensure a Chromium-based browser is open with an incognito/private window showing
<spec url>. Reuse an incognito/private window that is already there; otherwise
open a new one and navigate to that URL. Make the window roughly <viewport> or
larger.

If a sign-in page appears at any point, sign in with these test credentials:
<credentials from spec, if present>

Form hygiene: if a browser autofill or password-manager dropdown appears over a
form, press Escape to dismiss it before clicking. After clicking a form field,
confirm the caret is in that field. After typing into a field, visually confirm
it contains exactly the expected text before moving on. If it does not, select
all and retype.

Follow the Task steps in order. Judge strictly against the Verification rules:
a missing error message or a hung spinner is a failure even if nothing crashed.
Quote on-screen error text verbatim in failure reports.

You are executing this behavioral QA test:

<paste the full qa/*.md spec contents here>

You have <timeout_s> seconds. When done, or when a Verification rule is
conclusively violated, answer with a concise QA report:

VERDICT: PASSED or VERDICT: FAILED
Steps performed
Failures with exact on-screen text
Notes

The first line of the report must be exactly VERDICT: PASSED or VERDICT: FAILED.

Demo discipline

Implement requested product changes before searching for unrelated known bugs. Let Holo QA surface regressions from behavior. Do not edit files in qa/ to make a failure pass.

이 저장소의 다른 Skills

같은 저장소

expense-report-automation

hcompai/holo-desktop-cli

Read a folder of receipt images/PDFs, append rows to a LibreOffice Calc bookkeeping ledger (description, category, type, amount), save, and start a Mail draft with the sheet attached.

2026-06-1936

holo-desktop

hcompai/holo-desktop-cli

Sub-agent that binds to one OS window on the user's machine and drives it in the background via H Company's Holo3 VLM. If direct Holo tools are not visible, search for holo_desktop tools. In NemoClaw, call holo_desktop_launch once, then poll the same run_id until it finishes, passing attached images/files as media_paths.

2026-06-1936

apple-calendar

hcompai/holo-desktop-cli

Creating, editing, and managing events, invites, and recurring schedules in the Apple Calendar desktop app.

2026-06-1936

apple-contacts

hcompai/holo-desktop-cli

Finding, creating, editing, and organizing contacts and groups in the Apple Contacts desktop app, including linked cards from multiple accounts.

2026-06-1936

apple-mail

hcompai/holo-desktop-cli

Composing, sending, replying, searching, and triaging messages in the Apple Mail desktop app.

2026-06-1936

apple-messages

hcompai/holo-desktop-cli

Sending iMessages, SMS, and replies, plus reactions and search, in the Apple Messages desktop app.

2026-06-1936

name	holo-qa
description	Use when asked to QA Nimbus Desk, run a qa/*.md behavioral spec, verify UI behavior, check a regression, or confirm a UI fix. Drives the running app through the root Holo MCP tool, holo_desktop(task), and returns a QA report.

holo-qa: behavioral QA through root Holo MCP

Workflow

Ensure the Nimbus Desk dev server is running on http://localhost:5173. If it is not, start npm run dev in the background from this directory and wait until curl -s -o /dev/null -w "%{http_code}" http://localhost:5173 returns 200.
Pick the closest spec from qa/*.md. If the user names a spec, use exactly that file. Otherwise map the request by filename and H1 title: chat-* covers the support chat widget, auth-* covers login/session behavior, and tickets-* covers the ticket list.
Read the full markdown spec, including YAML frontmatter. Use the url, viewport, timeout_s, and credentials fields when composing the Holo task.
Call holo_desktop(task) once with a self-contained task using the template below. Wait for the final result.
Treat the Holo report as the UI evidence. Do not manually browse the app as a substitute. Quote failures verbatim.
If asked to fix a failed flow, fix the app code, then rerun the same spec once. If the rerun still fails, stop and summarize both reports.

Holo task template

You are a meticulous read-only black-box QA tester for Nimbus Desk.
Judge only what is visible on screen. Do not open developer tools, read source
code, inspect the DOM, or fix anything. Your job ends at observing and
reporting.

Ensure a Chromium-based browser is open with an incognito/private window showing
<spec url>. Reuse an incognito/private window that is already there; otherwise
open a new one and navigate to that URL. Make the window roughly <viewport> or
larger.

If a sign-in page appears at any point, sign in with these test credentials:
<credentials from spec, if present>

Form hygiene: if a browser autofill or password-manager dropdown appears over a
form, press Escape to dismiss it before clicking. After clicking a form field,
confirm the caret is in that field. After typing into a field, visually confirm
it contains exactly the expected text before moving on. If it does not, select
all and retype.

Follow the Task steps in order. Judge strictly against the Verification rules:
a missing error message or a hung spinner is a failure even if nothing crashed.
Quote on-screen error text verbatim in failure reports.

You are executing this behavioral QA test:

<paste the full qa/*.md spec contents here>

You have <timeout_s> seconds. When done, or when a Verification rule is
conclusively violated, answer with a concise QA report:

VERDICT: PASSED or VERDICT: FAILED
Steps performed
Failures with exact on-screen text
Notes

The first line of the report must be exactly VERDICT: PASSED or VERDICT: FAILED.

Demo discipline

Implement requested product changes before searching for unrelated known bugs. Let Holo QA surface regressions from behavior. Do not edit files in qa/ to make a failure pass.