1. Analyze the task and select the optimal LLM model and agent.
2. Dispatch the implementation sub-agent with a structured prompt.
3. Dispatch the judge agent to verify the implementation.
4. Parse the judge's verdict, score, and identified issues.
5. If the score is below the threshold and retries are available, retry with feedback.
6. If the score is above the threshold or max retries are exceeded, report the final results.
7. If max retries are exceeded, escalate to the user for a decision.
SynapticSkills