The Technical Verifier is a Claude Code subagent that checks an article's technical claims (commands, code, pricing, UI paths) against current official documentation before publication. It uses Perplexity for research and Firecrawl to scrape official sources, then produces a verification report with issues and suggested fixes, without auto-editing the article.
What you will be able to do
- Build a Claude Code subagent (technical-verifier) that extracts technical claims from a draft article.
- Run /verify against a draft to get a report of claims checked, issues found, and verified-accurate sections.
- Catch outdated pricing, deprecated model names, and wrong plan limits before publishing them.
- Choose a verification depth level (Quick, Standard, Deep) matched to how critical the article is.
- Fit verification into a pre-publish workflow: draft, review, verify, fix, re-verify, finalize.
Before you start
- Claude Code CLI installed (curl -fsSL https://claude.ai/install.sh | bash)
- Free Perplexity API key and free Firecrawl API key
- MCP servers for Perplexity and Firecrawl configured in Claude Code
- Agent folder structure (.claude/agents/) and research profile files set up from Article 1 of the series
Reference
| Setting / Command | Value | Notes |
|---|---|---|
| Command | /verify <path-to-article.md> | Runs the verification agent against a draft |
| Depth: Quick | 30s, major claims only | Quick sanity check |
| Depth: Standard (default) | 90s, all explicit claims | Normal publication |
| Depth: Deep | 3 min, claims plus implied assumptions | Use for flagship or critical tutorials |
| Agent tools | Read, Glob, Grep, Write, mcp__perplexity__search, mcp__perplexity__reason, mcp__firecrawl__firecrawl_scrape | Declared in the subagent's frontmatter |
| Agent model | sonnet | Set in subagent frontmatter |
| Report fields | Claims Checked, Issues Found (critical/minor), Verified Accurate, Publish Status | Output structure of the verification report |
| Cost per verification | About $0.20 | From the author's ROI breakdown |
Common errors and fixes
| What Goes Wrong | The Fix |
|---|---|
| Article references a retired model name (e.g. "GPT-4") that no longer appears in current pricing | Replace with the current model name and current price, sourced from the official pricing page |
| Article states a cost multiplier that's now wrong (e.g. "Opus costs 3x more than Sonnet") | Update to the current multiplier calculated from current official pricing, with the source cited |
| Article lists a tool's plan or execution limits (e.g. n8n tier limits) that have since changed | Correct the numbers to match current plan pages before publishing |
| Content relies on the LLM's own training data instead of live research, so it confidently states outdated info | Verify claims against Perplexity search results and Firecrawl-scraped official docs, not model memory |
| A critical pricing error gets treated the same as a minor nitpick | Apply human judgment to the report since the agent never auto-fixes, it only flags critical versus minor issues |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of PubFlow OS Agents: Build Your AI Research Team.
More in this section
- Lesson 4: The Agent That Replaced My $200/Month Competitor Analysis Tool
- Lesson 6: My Agent Found 47 Content Opportunities (Gap Analyzer Build)
Continue the course
Browse all lessons in the PubFlow OS Agents: Build Your AI Research Team course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
