技能 2026-08-17 09:54 UTC
langchain-agent-skills
由提交人 Szymon-Budziak 提交,主要更新类型为:依赖升级与工程维护。提交说明:chore: rebrand to Soba Labs, and make the skills work in Codex and OpenCode (#3)。更新了依赖与构建配置,跟进上游版本。
Commit提交信息原文
chore: rebrand to Soba Labs, and make the skills work in Codex and OpenCode (#3) * chore: rebrand marketplace, licence and README to Soba Labs The repository moved to the soba-labs org but its contents still pointed at Lubu Labs. GitHub 301-redirects the old paths, so nothing was broken by the move itself, but three things were wrong. - The marketplace manifest was named lubulabs-langchain-agent-skills, with a Lubu Labs owner, a lubulabs.com email and URL. - The README's install command said @lubu-labs-langchain-agent-skills, which did NOT match the manifest name (no hyphens), so that command already failed before this rename. It also told users to install an individual skill, while the installable units are the three plugin bundles. - LICENSE carried a Lubu Labs copyright. Marketplace renamed to soba-labs-langchain-agent-skills. That is a breaking change for anyone who already added the marketplace under the old name, hence the major version bump; they need to remove and re-add it. Copyright now names the legal entity. * feat: make the skills discoverable and installable in Codex and OpenCode The skills were already harness-agnostic in FORMAT: one SKILL.md with name and description frontmatter plus optional scripts/references/assets is exactly what Claude Code, Codex and OpenCode each read, and all three use progressive disclosure. Nothing in skills/ needed changing. What was missing was discovery and distribution for anything other than Claude Code. - .codex-plugin/plugin.json makes this a valid Codex plugin. Codex's skills-only plugin format is a manifest plus a skills/ directory, which this repo already had, so the manifest is the whole gap. - .claude/skills and .agents/skills symlink to skills/, so all three assistants discover them when this repo is the working project. OpenCode reads both paths, so the two links cover the three harnesses. Also fixes two wrong instructions in the Codex section of the README: - '$skill-installer install ... from <repo>' does not appear anywhere in OpenAI's documentation. Replaced with the documented directory paths. - It told users to copy skills into ~/.codex/skills. Codex's documented global skills directory is ~/.agents/skills; ~/.codex holds AGENTS.md, not skills. That path would silently do nothing. Adds a Harness Compatibility section stating the read paths per assistant.