What this is
A Chinese independent developer recently open-sourced a tool called Usora, targeting something most AI users encounter but no one systematically articulates—the experience generated through human-AI collaboration is being massively lost. The problem it wants to fix is concrete: you spend an hour today working through a complex problem with AI; next week, on a different project, you have to explain everything again—background, rules, judgment criteria, output format, all from scratch.
The author argues that this "collaboration experience" is now approaching a form of "capability"—not a Prompt (an instruction to AI), but a methodology that can be re-invoked by AI Agents (AI assistants that execute tasks autonomously). Usora aims to extract these methods from conversation history and crystallize them into reusable Skills. Its core mechanism is called Foundry: observe the collaboration process → collect activities → distill candidate skills → human verification → publish to a personal skill library.
Another design priority is cross-AI compatibility. Rather than binding to one of Codex (OpenAI's coding assistant), Claude Code (Anthropic's coding assistant), or Kimi (Moonshot AI's assistant), Usora acts as a universal Hub, letting users' "skills" transfer across different AIs. This responds to a real problem many face: today this AI, tomorrow that one—but capabilities can't follow.
Industry view
Supporters argue this direction hits a genuine pain point. In the past, when we used software, experience was captured in documents and SOPs (Standard Operating Procedures); now human-AI collaboration is growing rapidly, but "how to retain collaboration experience" is something almost no one handles. Usora's value lies in being the first to systematically propose both a methodology and a tool prototype.
Skepticism is also substantial. First, whether an open-source personal project can sustain maintenance is a real question—GitHub is littered with similar-idea tools that died mid-build. Second, the author himself admits in his write-up: if the filtering mechanism isn't well-built, the skill library becomes "another dump." Auto-converting every conversation into a Skill drowns signal in noise; manual filtering kills efficiency. Third, and more realistically: big players are already doing similar things—Anthropic is pushing Claude Skills, OpenAI builds project-level context in Codex, Cursor is also building team knowledge bases. An individual open-source project competing with this infrastructure has a narrow landing zone.
Our editorial judgment: the real difficulty here isn't technical—it's "what's worth capturing." That balance point hasn't been found by anyone yet; it will require either product design cleverness or ecosystem-driven evolution.
Impact on regular people
For enterprise IT: If your team is already using AI coding or writing tools at scale, Usora's approach is worth watching—it surfaces a long-term issue: the more you use AI, the easier single-point experience is lost, and team knowledge assets are being diluted.
For individual careers: Right now, using AI to write reports, build spreadsheets, or draft copy means reinventing the wheel each time. Once these tools mature, they could save significant time, but in the short term they're still far from typical workplace users.
For the consumer market: At this stage it mainly serves developers; regular consumers won't notice in the short term. But the idea of "AI remembering your preferences and how you work," once mature, will extend to everyday office work, consumer scenarios, customer service, and beyond.