Verified Task Implementation FAQs

Question 1

How does the LLM-as-Judge verification work?

Accepted Answer

The skill launches a specialized judge agent after implementation steps to evaluate artifacts against a 5.0 scale. If the score falls below the threshold, it triggers an automated fix-and-verify loop.

Question 2

Can I intervene during the automated process?

Accepted Answer

Yes, using the --human-in-the-loop flag allows you to set mandatory checkpoints where the agent pauses for your feedback and approval before moving to the next step.

Question 3

What is the benefit of the --refine mode?

Accepted Answer

Refine mode detects manual changes in your git workspace and maps them to specific task steps, re-verifying only the necessary parts of the workflow to ensure consistency.

Question 4

Why does the skill emphasize being an 'Orchestrator Only'?

Accepted Answer

By delegating file reading and writing to sub-agents, the main orchestrator keeps its context window clean, preventing the 'forgetting' behavior common in long, complex AI coding sessions.

Question 5

How does this skill handle long-running tasks?

Accepted Answer

It uses a directory-based state management system (.specs/tasks/). Using the --continue flag allows the agent to pick up exactly where it left off, verifying the current state before resuming.

Verified Task Implementation

Key Features

Use Cases

Verified Task Implementation

Key Features

Use Cases