Lore Shares Technical Deep-Dive on /share: Building Shareable Links for Claude Code Transcripts
Key Takeaways
- ▸/share enables developers to create and share Claude Code transcripts via one-click shareable links, supporting collaboration workflows, multi-agent handoffs, and product observability
- ▸Lore's architecture evolved through three iterations to solve real-world constraints: skill-based approach (context inefficient), script/CLI approach (limited to local environments), and local MCP servers (supporting sandboxed cloud environments)
- ▸The feature is distributed as a plugin containing multiple complementary skills (/share, /read, /fork) that enhance the Claude Code development experience and extend transcript utility
Summary
Lore has published a comprehensive technical breakdown of how its /share feature enables users to create and distribute shareable links from Claude Code session transcripts. Originally conceived as part of Lore's founding vision as a social network for AI sessions, /share has evolved into a critical infrastructure tool for developers collaborating on AI-assisted projects. The feature architecture consists of three primary layers: a user-facing skill wrapped in a plugin, a local MCP server for on-device functionality (file reading, clipboard management), and a remote MCP server orchestrating the underlying API. The feature supports diverse use cases including developer handoffs to colleagues or agents, product observability for Claude-based tools, and seamless integration across Claude's ecosystem (Code, Desktop, Cowork).
Lore's engineering journey reveals how real-world constraints shaped the architecture. Early iterations attempted to have agents read their own session files, but this proved inefficient—doubling context usage and breaking agent session compaction. The team iterated through multiple approaches: from direct agent-based reading, to local scripts and CLI tooling, eventually settling on local MCP servers to overcome sandboxing limitations in Claude Cowork. The final design elegantly bridges local and remote infrastructure, handles authentication transparently, and automatically copies shareable links to users' clipboards. The plugin also bundles complementary skills like /read (for querying existing threads) and /fork (for intent-based thread summarization).
- Local MCP servers running with user-level permissions bridge sandboxed cloud environments (Claude Cowork) and local filesystems, with built-in authentication flows to ensure secure access
Editorial Opinion
Lore's technical approach to /share demonstrates mature infrastructure design for collaborative AI development. By solving the real challenge of persisting and sharing AI session context, Lore is building essential tooling for a new class of AI-assisted workflows. The thoughtful evolution through architectural constraints—from context efficiency to sandboxing—shows the depth required in AI infrastructure, and open-sourcing their skill lowers barriers for similar innovations across the Claude ecosystem.
