- Lore started as a social network for AI prompts and sessions, evolving away from that but keeping the /share command useful for development and early users.
- The /share command is used to attach Lore links to pull requests, hand off work between users or agents, and provide observability for user-built skills and plugins.
- The command involves challenges like finding the right file, reconciling repeated shares, authentication, copying links automatically, and allowing custom titles and highlights.
- Architecture consists of three layers: a plugin-wrapped skill, a local MCP server for on-device functions, and a remote MCP server orchestrating the API.
- The skill initially read session files directly, causing context issues, so it was improved with a script, then a CLI, and finally a local MCP server for broader compatibility.
- A plugin bundles the skill and local server, namespacing commands like /lore:share, and adds /read and /fork skills for specific summarization.
- Local MCP server handles file reading, uploads, clipboard copying, and authentication using long-lived Upload API Keys instead of OAuth tokens for reliability.
- Remote MCP server abstracts 11 steps including S3 uploads, thread management, parsing, and analytics, making the plugin simpler and easier to maintain.
- The lore-plugin repo on GitHub contains skills, local MCP server source, and manifests for Claude, Codex, and experimental Amp support.
- Lore believes AI-native work introduces new primitives like threads, skills, and artifacts, and aims to build tools for frontier knowledge workers.