Best way to automatically check and fix errors after execute phase

[marcorobiati]

Can the Discuss-Plan-Execute-Verify cycle be specifically adapted to include a dedicated error-checking and fixing phase after execution-phase completes?

I encountered a situation where the system signaled the 'execute-phase' as complete and all tests as passing. However, upon moving to the interactive 'verify-work' phase, the application failed to build due to TypeScript errors and wouldn't load. Despite these critical failures, the verification prompts focused on minor UI details, such as the dark theme toggle.

Am I missing a standard procedure for this, or would implementing a formal Error-Checking/Fixing Cycle between 'execute-phase' and 'verify-work' be the best way to resolve these discrepancies?

I told the issues in the 'verify-work' and Claude started fixing things for 20 minutes but i'm not sure if it respected all previous work done by GSD

[trek-e]

This is a known pain point. Here's what's happening and how to handle it:

Why verification missed the build errors

The verifier agent checks against the phase goal and plan requirements, not against "does the app build and run." If the executor's tests passed (unit tests can pass even when the full app doesn't build — e.g., TypeScript type errors in files not covered by tests), the verifier sees green tests and moves on.

What to do now

1. Run /gsd-plan-phase --gaps — this reads VERIFICATION.md and creates targeted plans for the gaps.
2. Add a build check to your discuss-phase preferences — in your next /gsd-discuss-phase, mention that you want the executor to run a full build (npm run build / tsc --noEmit) as a post-execution gate, not just tests. This gets captured in CONTEXT.md and the planner will include it.

Preventing this in the future

The most reliable approach is to add build verification to your project's test suite. If npm test includes a TypeScript compilation check, the executor can't mark a plan as complete with type errors:

// package.json
"scripts": {
  "test": "tsc --noEmit && vitest run"
}

This way the executor's own test step catches build failures before verification even runs.

See User Guide for the full phase execution flow.