anthropics
Research-First Workflow Enforcement
Research-First Workflow Enforcement
Priority: HIGH
Problem: Claude Code implements before understanding. Observed pattern: try commands → fail → research docs → try more commands. Should be: research docs → understand → plan → implement.
Current Behavior:
Line 26-100 of transcript: immediate EAS credential commands Line 410: first documentation search (after many failures) Backward workflow wastes time Expected Behavior:
For complex unfamiliar tasks, offer: "This looks complex. Should I research [topic] first?" Don't jump to implementation for new problem domains Read relevant documentation before trying commands Show what was learned from research before proposing approach Impact:
Many failed attempts that could be avoided Solutions based on incomplete understanding User has to redirect AI to read docs Suggested Implementation:
Detect complex tasks involving new technologies/APIs Auto-suggest research phase: "I should review the EAS credential documentation before attempting setup. OK?" Show research findings before implementation Let user approve approach after research Example of Good Flow:
User: "Set up EAS credentials for iOS internal distribution"
AI: "This is a complex credential setup task. Let me first research the EAS documentation for iOS credential requirements, then propose an approach. OK?"
[After research]
AI: "I found that EAS requires either:
- Interactive credential setup with Apple ID, OR
- Pre-existing credentials configured on EAS servers, OR
- Local credentials (requires .p12 + .mobileprovision files)
Based on available environment variables [lists them], approach #2 seems most feasible. Should I proceed with checking what credentials are already configured on EAS?"
