Overview
Thegsd:verify-work command validates built features through conversational testing with persistent state. It confirms that what Claude built actually works from your perspective—one test at a time, with plain text responses, no interrogation.
Usage
Parameters
The phase number to verify (e.g., 1, 2, 3).
- If provided: Tests the specific phase
- If not provided: Checks for active sessions or prompts for phase selection
How It Works
The command follows a conversational testing workflow:- Session Management - Creates or resumes testing session
- Test Presentation - Shows one test case at a time
- Your Response - You describe what happened (pass/fail/issue)
- Diagnosis (if needed) - Automatically investigates issues
- Fix Planning - Creates verified fix plans ready for execution
- State Tracking - Records all results in UAT.md
The testing is conversational—not a checklist. You describe what you see in plain language.
Conversational Testing
Unlike traditional UAT, this is a dialogue: System: “Try logging in with an invalid password. What happens?” You: “I see a generic error message, but it doesn’t say whether the email or password was wrong” System: Investigates code, diagnoses issue, creates fix planOne test at a time. Plain text responses. No form filling or rigid pass/fail checkboxes.
Automatic Issue Handling
When you report an issue:- Diagnosis - The system investigates the codebase
- Root Cause - Identifies what’s wrong
- Fix Plan - Creates a gap closure plan
- Verification - Validates the fix plan is executable
- Ready for Execution - Marked for
--gaps-onlymode
Issue to Fix Flow
Persistent State
Testing sessions persist across conversations:- Resume testing where you left off
- Review previous test results
- Track fixes across multiple cycles
Files Created/Modified
Tracks all test results, issues found, and fix status for the phase
Gap closure plans created when issues are found (marked with
gap_closure: true in frontmatter)Examples
Verify Specific Phase
Resume Active Session
Complete Verification Cycle
Testing Best Practices
Be Specific About Issues
Good: “The error message says ‘Invalid credentials’ but doesn’t indicate whether it’s the email or password that’s wrong” Less Helpful: “It doesn’t work”Test Real Scenarios
Good: “I tried uploading a 10MB file and got a timeout after 30 seconds” Less Helpful: “File upload seems slow”Report Unexpected Behavior
Good: “After deleting an item, the list count still shows the old number until I refresh” Less Helpful: “Deleting is buggy”The more specific you are, the better the system can diagnose and fix issues.
What Gets Tested
The system generates test cases based on:- Phase goals from ROADMAP.md
- Implementation details from PLAN.md
- User scenarios from REQUIREMENTS.md
- Edge cases identified during planning
Success Criteria
Verification succeeds when:- All test cases have been executed
- Issues have been documented and diagnosed
- Fix plans created for all identified gaps
- You’re satisfied with the quality
You don’t need to find zero issues. The goal is to identify what needs fixing, create plans, and iterate.
Workflow Gates
All workflow gates are preserved:- Session Management - Resume capability and state tracking
- Test Presentation - One test at a time, conversational format
- Diagnosis - Automatic investigation of reported issues
- Fix Planning - Verified gap closure plans
- Routing - Next-step suggestions
Next Steps
After verification: If gaps found:Allowed Tools
This command has access to:- Read - Read plans, code, and test files
- Bash - Execute tests and commands
- Glob - Find relevant files
- Grep - Search codebase for issues
- Edit - Modify UAT.md with results
- Write - Create gap closure plans
- Task - Delegate diagnosis to specialized agents