Download Latest Version
OpenSpace_ The Skill Management Layer for AI Agents source code.zip (99.0 MB)
Get an email when there's a new version of OpenSpace
| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| OpenSpace_ The Skill Management Layer for AI Agents source code.tar.gz | 2026-07-17 | 98.1 MB | |
| OpenSpace_ The Skill Management Layer for AI Agents source code.zip | 2026-07-17 | 99.0 MB | |
| README.md | 2026-07-17 | 2.6 kB | |
| Totals: 3 Items | 197.1 MB | 1 | |
OpenSpace v2 turns OpenSpace into a quality-first Skill Hub for AI agents.
This release focuses on a simple idea: agents should not only save and reuse skills, they should know which skills actually work after real tasks. OpenSpace v2 adds quality records, controlled skill evolution, cloud skill browsing, and a stronger runtime harness around agent execution.
Highlights
- Quality-first Skill Hub: skills are organized around real task outcomes, reuse signals, lineage, and reviewable context.
- Skill Wiki visualization: the new Skill Wiki view shows package-tree browsing and in-package skill search, making shared skills easier to inspect.
- Controlled skill evolution: runtime evidence can lead to FIX, DERIVED, or CAPTURED skill updates, with review and validation before active skills are replaced.
- Skill quality records: OpenSpace tracks whether skills were selected, applied, completed work, failed, or fell back.
- Local-first workflow: agents keep running and reusing skills locally, while cloud surfaces help people browse and review shared skills.
- Stronger agent harness: CLI, Python API, MCP, gateway, dashboard, tools, permissions, sessions, and quality evidence now share a clearer runtime boundary.
What's New
Skill Quality from Real Tasks
OpenSpace v2 records quality from actual agent work instead of relying only on skill text. It tracks task-result quality, tool reliability, fallback behavior, and evidence that helps decide whether a skill should be reused or improved.
Controlled Skill Evolution
Skill updates are now framed as evidence-driven changes. Broken or outdated skills can be fixed, useful variants can be derived, and successful workflows can be captured as reusable skills. Unclear signals can be held for review instead of being promoted automatically.
Skill Wiki and Package Browsing
The new Skill Wiki visualization presents shared skills as a searchable package tree. This connects the v2 Skill Hub story to the cloud browsing experience and makes skill collections easier to understand at a glance.
Runtime and Integration Improvements
OpenSpace v2 strengthens the execution layer used by CLI, Python API, MCP, gateway, dashboard, and tool runtime flows. Sessions, permissions, sandboxing, tool results, and quality evidence are treated as part of the same agent harness.
Documentation
- English README:
README.md - Chinese README:
README_zh.md - Host integration guide:
openspace/host_skills/README.md - Configuration guide:
openspace/config/README.md
