epic: Modern Agentic CLI — intent-based UX for AI agents

[padak]

Vision

kbagent is already a strong foundation for AI agents (structured JSON output, self-documentation via context, permission model, multi-project parallelism, Claude Code plugin). But to move from being a tool for agents to a partner for agents, we need higher-level primitives.

This epic tracks features that give agents pre-assembled context instead of forcing them to build it from individual API calls.

Child issues

• feat: kbagent diagnose — intent-based debugging for AI agents #147 — kbagent diagnose — intent-based debugging (related to kbagent check: Platform observability with AI-powered investigation #74)
• feat: kbagent job verify — post-execution validation #148 — kbagent job verify — post-execution validation
• feat: kbagent pipeline trace — end-to-end data flow mapping #149 — kbagent pipeline trace — end-to-end data flow mapping
• feat: kbagent recipe — workflow templates for common multi-step tasks #150 — kbagent recipe — workflow templates for multi-step tasks
• feat: kbagent suggest — contextual recommendations for AI agents #151 — kbagent suggest — contextual recommendations

Also related

• kbagent check: Platform observability with AI-powered investigation #74 — kbagent check / kbagent observe — platform observability with AI investigation

Design principles

1. Intent over CRUD — commands should match what agents want to do, not what APIs offer
2. Context pre-assembled — don't make agents correlate 5 API calls when one command can do it
3. Executable suggestions — output should include runnable next-step commands, not prose
4. Layered abstraction — low-level CRUD remains available, high-level commands build on top

Background

Inspired by the "agentic user experience" (AUX) discussion — headless doesn't mean brainless. The ergonomics of the agent interface matter more than API coverage.