team-harness
Team Harness
Kick off a Claude Code agent team with specialized teammates that work in parallel and communicate directly. Unlike feat-harness (sequential subagents), team-harness uses agent teams where teammates see errors, fix them, test, and loop — without waiting for a central orchestrator.
Why Teams Over Subagents
Subagents are fire-and-forget: they do work and report back. They can't talk to each other. If the QA agent finds a bug, it has to report back to you, then you dispatch the engineer again — that's slow.
Agent teams solve this: the QA teammate messages the engineer directly. The dev-environment teammate broadcasts error logs to everyone. The loop tightens from minutes to seconds.
Pre-flight
Before anything, verify agent teams are enabled:
# Check if the experimental flag is set
grep -r "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS" ~/.claude/settings.json 2>/dev/null
More from derek-x-wang/skills
life-os
>
10convex
Build full-stack TypeScript applications with Convex - a reactive backend-as-a-service with real-time database, serverless functions, and automatic caching. Use when working with Convex projects, writing queries/mutations/actions, defining schemas, using the Convex CLI, or integrating Convex with React/Next.js/Node.js applications.
4better-convex
ALWAYS use this skill when working with convex or better-convex. Covers the common end-to-end feature path using cRPC + ORM + auth + React, with setup/bootstrap and niche depth in references.
3manage-my-skills
Manage, update, and install personal agent skills from Derek-X-Wang/skills. Use when the user wants to create, edit, update, or install skills. Handles the full workflow — edit locally, commit, push, then npx skills install.
3cmux-team
Run a Claude Code agent team inside a cmux claude-teams session. Use when the current session was launched via `cmux claude-teams` and you want to spawn teammates that surface as separate cmux panels. Covers detection, the spawn → coordinate → shutdown contract, and a mode index — pick the matching `modes/<mode>.md` for the topology you actually want (feature build, AFK runner, parallel AFK runners, structured debate, parallel research).
3repo-knowledge-share
Cross-project knowledge sharing. Catalogs features and tech stacks across repos so you can reuse implementations. Use when building a feature that another project might already have, when the user asks "has any project done X?", or says "scan this project" / "update knowledge". Also triggers automatically on any invocation if the current project's knowledge is stale (>7 days since last scan).
3