Skip to content

README.md

Latest commit

 

History

History
34 lines (23 loc) · 1.67 KB

README.md

File metadata and controls

34 lines (23 loc) · 1.67 KB
Raw

qa/cli — QA a web UI via a CLI + Claude Code skill

Same SDK calls as qa/mcp, different interface. Where qa/mcp exposes the agent as an MCP server, this example exposes it as a shell command and uses a Claude Code skill to teach Claude when to invoke it.

CLI

uv run qa-cli review --url https://example.com --instruction "look for broken links"
uv run qa-cli visual --url https://example.com --question "what color is the heading?"
  • review prints a ReviewResult JSON on stdout (verdict, summary, findings, steps_taken).
  • visual prints a short free-text answer.
  • Progress logs go to stderr.

Skill

The skill lives at skills/hai-qa-via-cli/SKILL.md. Claude Code picks it up automatically and triggers when the user asks to QA a URL or get a quick visual answer about a page.

"QA https://example.com for accessibility issues."

Claude invokes the CLI via Bash, parses stdout, and surfaces findings in chat.

How it differs from qa/mcp

Aspect qa/mcp (MCP) qa/cli (CLI + skill)
Wiring .mcp.json registers an MCP server skills/hai-qa-via-cli/SKILL.md
Invocation in Claude Native mcp__... tool call Bash invocation of qa-cli
Result handling Structured tool return Claude parses stdout JSON
Runnable outside Claude Need an MCP client Just run the CLI
Server process Yes (FastMCP, stdio) No — one-shot subprocess

Same agent definition, same prompt, same answer_format. Choose the surface that fits — MCP when you want a long-lived tool surface in Claude Code; CLI when you want a thing humans and scripts can run too.