Six months ago, discussions about AI coding tools centered on "which model writes better code."
Now the trending AI coding projects on GitHub don't have model names in their titles. Instead, you see words like Skills, Switch, and OS.
The toolchain is growing the same skeleton.
Skills Become the New Capability Unit
Addy Osmani's agent-skills project, 37K stars, positioned as a "production-grade engineering skills library." Matt Pocock's Skills collection hit 60K stars. Anthropic's Agent Skills standard, launched last year, is now followed across the board by OpenAI, Microsoft, and Cursor.
Skills essentially extract "what this Agent can do" from inside the model and turn them into pluggable modules. You don't need to retrain the model to gain new capabilities - just install a Skill.
It's very similar to the logic of a phone's App Store. The model is the operating system, Skills are the apps.
But the Skills ecosystem is still in the "each platform does its own thing" phase. Claude Code Skills, Codex Skills, OpenClaw Skills - incompatible formats, different installation methods.
Cross-Tool Managers Arrived
cc-switch was born for this problem.
A desktop tool that centrally manages Claude Code, Codex, OpenCode, OpenClaw, and Gemini CLI - switching between them without opening different terminals and typing different commands each time. 65K stars, 7K weekly growth, the fastest-growing infrastructure project in the AI coding toolchain right now.
Another direction: xingkongliang's skills-manager, managing AI Agent Skills across 15+ coding tools. Only 1.3K stars currently, but the direction is right.
When using multiple coding Agents simultaneously becomes the norm (which is almost inevitable - different tools for different scenarios), cross-tool management stops being a nice-to-have convenience feature and becomes a necessity.
Agent OS: A Deeper Paradigm
Q00's ouroboros project proposes a more radical concept: "Stop prompting. Start specifying Agent behavior through specifications."
This approach transforms Agents from "you tell it what to do" to "you define what it is." Skills describe capabilities, OS describes behavioral norms, and together they form a reusable Agent definition.
If this path works out, future Agent development might look more like writing configuration files than prompt engineering.
What This Means for Developers
Short term: if you're using multiple coding Agents, start using tools like cc-switch to manage them now - the efficiency gain is noticeable.
Medium term: Skills format unification is highly likely. Anthropic started it, OpenAI and Microsoft are following, community projects are building bridges. Wait for a de facto standard to emerge, then migrate toward it.
Long term: when "defining Agents" replaces "prompting Agents," the competition in AI coding tools shifts from model capability to ecosystem richness - whoever has the bigger Skills marketplace and the better Agent OS wins.
This trajectory mirrors MCP protocol's development. MCP took half a year to go from experiment to industry standard. Skills standardization might be faster, since it directly impacts developers' daily work efficiency.
Related reading:
Main sources: