Summary
Implement expertise-level and interaction-mode system that adapts Zeph's behavior to the user's skill level and current task context. The agent should augment the developer, not replace them — different users need different interaction styles.
Motivation
Current system prompt is static and one-size-fits-all. This leads to:
- Over-explaining for experts (wastes tokens and time)
- Under-explaining for beginners (confusing, promotes vibecoding)
- No distinction between coding tasks, general Q&A, and automation
Design
Two orthogonal dimensions:
Expertise Level (beginner / intermediate / expert) — controls communication style:
- Beginner: explains every step, asks for understanding confirmation
- Intermediate: challenger mode, shows trade-offs, asks probing questions
- Expert: minimal commentary, executes efficiently, flags only genuine risks
Interaction Mode (assistant / coding / automation) — controls focus and persona:
- Assistant: general-purpose helper, conversational
- Coding: developer tool, code-centric, strict conventions
- Automation: task executor, minimal interaction, maximum throughput
Architecture
See .local/plan/m27-interaction-modes.md for full design doc.
Issues
Dependency Graph
#591 (config types)
├── #592 (prompt templates) ──┐
├── #595 (CLI flags) │
├── #596 (--init wizard) │
└───────────────────────── #593 (refactor build_system_prompt)
└── #594 (runtime switching)
Summary
Implement expertise-level and interaction-mode system that adapts Zeph's behavior to the user's skill level and current task context. The agent should augment the developer, not replace them — different users need different interaction styles.
Motivation
Current system prompt is static and one-size-fits-all. This leads to:
Design
Two orthogonal dimensions:
Expertise Level (
beginner/intermediate/expert) — controls communication style:Interaction Mode (
assistant/coding/automation) — controls focus and persona:Architecture
See
.local/plan/m27-interaction-modes.mdfor full design doc.Issues
Dependency Graph