QA Swarm
Launches 3-5 QA subagents in parallel against the current project, each auditing a distinct dimension, then synthesises findings into one report ranked by severity.
Entry verified April 21, 2026
The short answer
Three specialists run side-by-side: an SEO agent (H1s, meta descriptions, schema, robots, internal linking); a content quality agent (thin content, placeholders, Irish specificity, editorial verdicts); a UX/data integrity agent (navigation, dead links, demo pages, vendor data consistency). Output ranks CRITICAL > HIGH > MEDIUM > LOW with a Quick Wins section.
When to use it
Run after a content batch lands, monthly against evergreen pages, or before a major rebrand. Each agent returns a prioritised list independently, so the synthesised report catches issues that siloed checks would miss.
Setup
- 1
Save the file as ~/.claude/commands/qa.md.
- 2
Invoke: /qa against the current project — the skill dispatches the agents in parallel.
- 3
Each agent returns a prioritised list, which the skill synthesises into one combined report.
- 4
Final report is ranked CRITICAL > HIGH > MEDIUM > LOW, with a Quick Wins section of fixes under 30 minutes each.
- 5
Spawning 3-5 agents multiplies token spend — scope the project path tightly before running.
Example
You: /qa pages/claude-code-mcp-stack.js Claude: 5 agents dispatched. Synthesis: 2 P1s (stale link, missing schema), 4 P2s. Full report below.
Source & attribution
- Author
- Bryan Collins
- Licence
- MIT (author's own work)
- Type
- Original
Original pattern published under MIT — attribution preserved by convention, not licence requirement.
Caveats
Costs 5x tokens per run. Budget accordingly.
56 skills across 10 categories, all licence-verified.