Using Expert Mode

Purpose

Choose the smallest correct workflow for the user's request. Use skills to structure substantive work, not to add ceremony to trivial interactions.

Core Rule

Before substantive work, follow this flow:

1. Identify whether the request starts a workflow
2. Load the most relevant skill
3. Follow that skill until you reach a natural handoff or completion point

For trivial interactions, respond directly.

What Counts as Substantive Work

Load a skill before:

• planning or design work
• implementing a feature or refactor
• debugging a bug or failing test
• reviewing completed work
• using external libraries, frameworks, or APIs where current docs matter

Usually respond directly for:

• short acknowledgements
• one-off clarification questions
• simple factual answers that do not begin a workflow

Skill Selection

• brainstorming: explore a design before implementation
• writing-plans: turn approved requirements into an implementation plan
• executing-plans: carry out an existing plan task by task
• systematic-debugging: investigate bugs or failing tests root-cause-first
• test-driven-development: implement behavior starting from a failing test
• completing-work: verify status before claiming success
• context7-mcp: fetch current official documentation

Load only what you need now. If the task changes shape, load the next relevant skill at that point.

Stop Conditions

• If requirements are still unclear, ask the clarifying question before loading more skills
• If a workflow is already in progress, do not reload the same skill just to continue normal execution
• If invoking a skill would add no value, do not do it

Red Flags

• Loading a skill just to ask a one-line clarification
• Re-invoking the same skill every turn
• Treating skills as ritual rather than guidance
• Starting implementation, debugging, or review work without selecting a workflow