Advanced Usage

This chapter covers model usage inspection, cache and context diagnostics, manual intervention during execution, and custom tools or SKILLs for reusable team workflows.

【Image: Advanced usage overview, labeling the usage panel, context editing entry, tool editor, and SKILL editor】

What This Chapter Contains

Page	Use it when you want to
Usage and Model Performance	Inspect token usage, cache hits, context pressure, speed, and jitter.
Manual Intervention and Context Editing	Pause, redirect, add constraints, or edit global and local context while Agent is working.
Custom Tools and Custom SKILLs	Package team scripts, checks, and repeatable workflows into reusable Agent capabilities.

Suggested Troubleshooting Order

When a task behaves poorly, check it in this order:

1. Inspect usage and model performance to see whether the issue is context size, cache misses, model tier selection, or network jitter.
2. Inspect execution behavior to see whether Agent misunderstood the task, picked the wrong tool, or needs sharper constraints.
3. Package the workflow if it repeats often, so you do not have to rewrite the same prompt each time.

【Image: Advanced troubleshooting flow from usage panel to context edits to tool and SKILL packaging】

Common Scenarios

Symptom	Read first
A task suddenly consumes many tokens	Usage and Model Performance
Similar prompts have low cache hits	Usage and Model Performance
First-token or total latency jitters	Usage and Model Performance
Agent is already running in the wrong direction	Manual Intervention and Context Editing
You want to change constraints for this task only	Manual Intervention and Context Editing
You want all future tasks to follow team rules	Manual Intervention and Context Editing
You want Agent to call an internal script	Custom Tools and Custom SKILLs
You want to reuse a repeatable workflow	Custom Tools and Custom SKILLs

Relationship to Basic Configuration

Basic configuration controls Agent defaults. Advanced usage helps you observe and intervene after tasks start.

Basic setting	How to tune it later
Pressure token threshold	Compare it with Pressure in the usage panel.
Review rule	Decide which plans, tools, and SKILLs should require confirmation.
Calling mode	Compare actual high-quality, lightweight, and vision model consumption.
Disable Tools / AI review	Decide which custom capabilities are safe to expose to Agent.

Next Step

Start with Usage and Model Performance to understand task cost and stability.