💡 Feature Request: System Prompt Refactor — Per-Mode Tools, Mode Visibility, and Prompt Slimming

[shariqriazz]

Hey folks,

I've been working on a major refactor of how the system prompt is generated. The current structure works, but it's bloated and too rigid once you start building real multi-agent workflows.

---

🧱 Current Prompt Structure:

Right now, every mode (built-in or custom) inherits a shared structure like this:

1. Role
2. Custom Instructions
3. Objective
4. Tool Usage Guidelines
5. Capabilities
6. Tools (group-based, not granular)
7. Other Available Modes (auto-injected)

---

🔻 Key Issues:

1. Tools are grouped, not granular

You could uncheck groups like "Use Browser", but you couldn’t assign individual tools to specific modes. Every tool in a group was all or nothing.

2. Hardcoded Objective / Capabilities / Guidelines

These were globally injected into every mode — even if the mode didn’t use tools or had a different purpose. There was no way to override or slim them down per mode.

3. Mode visibility was global

Every mode (even custom ones) was automatically added to the "Other Available Modes" section of every prompt — no control, no filtering.

---

✅ Refactor Details:

🛠️ 1. Per-Mode Tool Selection

Each mode now selects individual tools it needs — no more depending on hardcoded groups.
If a tool is enabled, its usage instructions and capabilities are pulled into the prompt.
If it's not, it's completely excluded.

🧼 Tools now define their own behavior. If they're not in use, they're gone.

---

✏️ 2. Custom Objectives + Capabilities in Instructions

The old globally shared objectives, capabilities, and tool usage guidelines are now removed.
Modes now define their own objective and purpose using their Custom Instructions field.

🎯 Modes are now truly independent agents — each with its own purpose and language.

---

👻 3. Per-Mode Visibility (Other Available Modes)

Modes can now define exactly which other modes they want visible.
So if you want ask to only see orchestrator, you can do that.
No more auto-injection of everything.

🔒 This removes prompt clutter and prevents irrelevant mode-switching behavior.

---

📉 Results:

• ✅ Prompt size is ~30–50% shorter
• ✅ Token usage is tighter and more consistent
• ✅ No unnecessary tool clutter or duplicated blocks
• ✅ Cleaner orchestration logic
• ✅ Easier to scale, debug, and customize

This one change unlocks a ton of flexibility while making the prompt pipeline leaner and more maintainable.