Agent Engineering Hub

Context / Concept engineering and subagent patterns rising

Context / Concept engineering and subagent patterns rising

Key Questions

What is the shift from prompt engineering to context engineering?

Context engineering involves subagent/skills, token savings, self-improving agents, and patterns like Gemma 4 agentic workflows. It includes domain modeling and Goose OSS progressive disclosure. This rises with hierarchical prompts and CLAUDE.md routing.

What is SkillX?

SkillX automatically constructs skill knowledge bases for agents. It enables reusable, domain-specific capabilities. This advances context engineering.

How are skills shared in agent development?

Developers share skills, not just code, assuming code is cheap; skills like SureThing 2.0 are personalized. @svpino highlights this trend. It supports token-efficient, self-improving agents.

What is Gemma 4's agentic capability?

Gemma 4 is multimodal (256K context), open-weight, deployable on Pi/GCP, with agentic workflows guide. Google's release emphasizes no-strings-attached use. It powers context-engineered agents.

What is AgentBuilder?

AgentBuilder builds AI agents from plain English YAML, simplifying creation. It introduces no-code agent development. This fits subagent patterns in context engineering.

What is Agentic-MME?

Agentic-MME evaluates what agentic capabilities add to multimodal intelligence. It's a benchmark paper. It underscores rising subagent patterns.

What is Meta-Harness?

Meta-Harness optimizes model harnesses end-to-end, auto-engineering agent setups overnight like AutoAgent. It enables self-optimizing systems. This advances context engineering.

How does LangGraph improve AI agents?

LangGraph uses prompting for smarter AI agents and workflows. It supports hierarchical and reusable patterns. Videos demonstrate building with it for context engineering.

提示工程→上下文工程:子agent/技能、token节省、自improving、Gemma 4 agentic(multimodal 256K/open-weight/Pi/GCP)、SKILL0 RL、Agentic-MME;领域建模、Goose OSS progressive disclosure/skills;Harness、AgentBuilder YAML、hierarchical prompts、@rubenhassid Claude setups、@svpino skills(SureThing 2.0);SkillX auto-skill;CLAUDE.md routing。新证据:Goose Managerbot、@rubenhassid、Gemma 4、Meta-Harness/AutoAgent、Claude Code、OpenClaw Skills、Spring AI、LangGraph、AgentBuilder、VS Code、Perplexity、@svpino、test-time adaptation、SkillX、marketing workflows。

Sources (33)
Updated Apr 8, 2026