Claude Workforce

Claude Workforce

*A skill tells an AI what to do. An employee knows what to do, refuses to do anything else, and can prove it did the job.*

**Supersedes** [claude-enforcer](https://github.com/odysseyalive/claude-enforcer). If you run the enforcer today, [here's the migration path](#coming-from-claude-enforcer).

**PAIRS WELL WITH:** playwright-mcp. Nobody here is hired without a way to prove the work is right, and the employees who deal with websites are the hardest to hire, because clicking around a web page rarely leaves behind a clean pass-or-fail signal the way finished code does. This tool records you logging into a site once, then turns that recording into a repeatable test the employee can actually run, so the proof is a test that passes, not a model's say-so. It also handles fetching web pages, which these employees otherwise can't do on their own. [Visit the repo →](https://github.com/odysseyalive/playwright-mcp)

Contents

• Not Every Brain Does Every Job
• How We Got Here
• Getting Started
  • Install
    • Into a specific Claude config directory
  • Updating
• Talking to the Company
• Three Tiers, Measured
• Which Model on Each Tier
• Employee Handbooks
• A Handbook Isn't Finished Until a Stranger Can Follow It
• When Something Goes Wrong, the Document Is at Fault
• Your CLAUDE.md Gets Deleted
• The Long Session Is the Expensive One
• The Honest Parts
• If Your Repo Has a Test Corpus
• Platform Facts Expire
• [Coming from Claude Enforc