AgentSkillsCN

beads-manual-qa

使用Agent Browser对Beads任务进行人工UI质量保证。验证流程、视觉效果与基本质量信号，随后将结果记录至Beads。适用于需要对UI变更进行验证时使用。

SKILL.md
--- frontmatter
name: beads-manual-qa
description: Manual UI QA for Beads tasks using Agent Browser. Verify flows, visuals, and basic quality signals, then write results to Beads. Use when UI changes need validation.
compatibility: Requires agent-browser CLI and Beads CLI (bd)
read_when:
  - manual UI QA
  - UI verification
  - UI smoke testing
  - verify UI changes
metadata:
  short-description: Manual UI verification using Agent Browser
  version: 0.1.0

Beads Manual QA

Mission

Verify UI changes with Agent Browser and record evidence and findings in Beads.

Orchestrator role

This skill is typically invoked by $beads-orchestrator. Users normally start with the orchestrator, not this role.

Inputs

  • •Task id
  • •Environment URL and credentials
  • •Scope: pages, flows, devices, and known risks
  • •Verification steps from the worker (if provided)

Rules

  • •Must use Agent Browser CLI for interaction and evidence capture.
  • •Do not change code or data beyond what is required to test.
  • •If the environment is unavailable, record a blocker in Beads and stop.

Tmux integration

  • •Run this skill from the Manual QA pane in the tmux session (scripts/tmux-orchestrator.sh start). Attach using scripts/tmux-orchestrator.sh attach and check the pane title to confirm you're in the correct task.

Procedure

  1. •Read the task and scope:
    • •bd show "$TASK_ID"
  2. •Establish the test environment:
    • •Use the worker's verification steps or the task description.
    • •If the URL or credentials are missing, record a blocker in Beads.
  3. •Start a QA session:
    bash
    agent-browser --session "qa-$TASK_ID" open "$URL"
agent-browser snapshot -i --json
  4. •Run verification modes (pick the ones relevant to scope):
    • •Critical user flows: navigate core journeys, fill forms, and verify UI state.
    • •Visual checks: capture screenshots for key states and compare with expected UI.
    • •Accessibility spot-check: run existing a11y tooling (axe/Lighthouse) if configured, otherwise do quick label/heading/focus checks.
    • •Quality audit: run Lighthouse if available to spot performance/best-practice regressions.
  5. •Capture evidence:
    bash
    agent-browser screenshot "qa-$TASK_ID-home.png"
agent-browser snapshot -i
  6. •Write results to Beads:
    • •Environment URL, steps executed, findings, evidence file names, and any blockers.
  7. •Close the session:
    • •agent-browser close