Show HN: Async multi-person collaboration skill for Claude Code

Collab Session Async collaborative thinking between multiple human+Claude pairs. Each participant works in their own Claude Code session, saving progress to a shared workspace. No merge conflicts by design — and the journey is preserved, not just the conclusion. How it works Two people (each with their own Claude Code session) work on the same topic asynchronously. Every /collab save writes a uniquely-named markdown file — nobody edits anyone else's files. When a colleague runs /collab join, all files are assembled into a narrative handoff brief so they can pick up where you left off. Simon + Claude Dan + Claude │ │ ├─ /collab save │ │ → simon_20260314T142001Z.md │ │ → git push │ │ │ │ /collab join ──────┤ │ ← git pull │ │ ← handoff brief │ (resolves [[wikilinks]] inline) │ │ │ /collab save ──┤ │ → dan_...Z.md │ │ → git push │ │ │ ├─ /collab refresh │ │ ← sees Dan's new block │ │ │ ├─ /collab compress │ journey-style _summary.md │ │ ├─ /collab reflect skill-project │ cross-session _patterns.md │ │ (only after 2+ closed sessions) On launch, a SessionStart hook nudges each participant about active sessions they've touched — no more "which session was I in?" at the start of the day. Why no vector DB, no RAG Cross-session recall is done via markdown primitives, not embeddings: [[session-NNN#phase-N]] wikilinks between summaries for graph navigation, /collab reflect to extract patterns from closed sessions into _patterns.md, and tiered retrieval in /collab catchup (compact index → timeline → full block) for bounded token cost. No vector DB, no index to rebuild, no model lock-in. If you'd rather run embeddings yourself, the folder is plain markdown — point any RAG tool at it. What this is good for Not just coding. Any topic two or more people want to think through together, asynchronously, with Claude as a partner in each seat. Research and analysis Two researchers reviewing papers on a topic, each with Claude pulling threads Market or competitor analysis where each person covers a segment Literature review split across a team, assembled into one journey summary Design and decision-making (pre-code) Co-designing an API, schema, or system architecture before anyone writes code Writing an RFC or design doc where each reviewer's reasoning is preserved, not just their verdict Weighing product direction or pricing options with the trade-offs captured Coding work where the why matters Architecture choices where future teammates need to know what was rejected and why Bug investigations where the dead ends are as valuable as the fix Refactoring decisions — the discussion that led to the final diff, preserved next to it What sets it apart for coding: a finished branch tells you what two colleagues agreed on. Collab Session captures why — the decisions, the disagreements, the approaches tried and abandoned. /collab compress writes a journey-style narrative with contributor attribution; /collab reflect surfaces patterns across closed sessions. Six months later, the next person on the codebase can read the lineage, not just the commit. Install claude plugin add AI-Collab-Skill/collab-session-skill Quick start # 1. Set your identity (once per machine) /collab whoami simon # 2. Set up a shared workspace (once per team) /collab init my-project # Choose: drive (network/cloud folder) or mini-repo (dedicated git repo) # 3. Start a session /collab new my-project api-design # Answer: goal, prior session? # 4. Brainstorm with Claude, then save /collab save #…

This excerpt is published under fair use for community discussion. Read the full article at GitHub.