team

Decompose tasks into parallel claude -p worker streams with judge-panel verification and synthesis.

Updated Jul 10, 2026
One-click install
npx skills add https://github.com/Kaleb-Rupe/aurora --skill team-kaleb-rupe
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: team
Source: https://github.com/Kaleb-Rupe/aurora/tree/main/claude/skills/team
Command: npx skills add https://github.com/Kaleb-Rupe/aurora --skill team-kaleb-rupe

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Complex tasks often exceed what a single agent pass can handle reliably, and naive parallelization risks overlapping work, leaked secrets, and unverified results. This Skill runs the Aurora orchestration engine to split a task into disjoint workstreams, verify them adversarially, and merge one synthesized answer. ## Core Features & Use Cases - Parallel decomposition: Classifies the task, splits it into pairwise-disjoint scopes, and fans out parallel claude -p workers. - Adversarial verification: Each workstream gets falsification review and mechanical secret-scrubbing, with a fail-closed judge panel at the join. - First-party sandboxing: Workers have no Bash, no web, no sub-agent spawning, all MCP denied, and no access to secrets; budget is capped at 500K tokens. - Use Case: Ask to have a team refactor a large codebase module — the engine decomposes it into independent scopes, runs workers in parallel, verifies each result, and returns one synthesized answer. ## Quick Start Type /team followed by your task, or ask to fan out a task across a team of parallel Claude workers, and relay the streamed progress and final synthesized answer.

Frequently Asked Questions about team

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a task with parallel Claude workers?

Invoke the /team command or ask to fan out a task, and the Skill calls the Aurora CLI with your task passed via a quoted heredoc on stdin. The engine decomposes the task, runs parallel claude -p workers, and streams progress plus a synthesized answer.

How does multi-agent orchestration verify worker results?

Each workstream undergoes adversarial falsification review and mechanical secret-scrubbing before a judge panel verifies results at the join. The panel is fail-closed, so unverified output does not reach the final synthesis.

Can parallel workers access the web, Bash, or MCP tools?

No. Workers run in a fail-closed environment with no Bash, no web access, no sub-agent spawning, and all MCP servers denied. No secrets reach the workers, and only your authenticated claude -p binary is used.

Why must the task be passed through a quoted heredoc?

The quoted 'AURORA_TASK' heredoc delimiter disables all shell expansion, so tasks containing quotes, $(), backticks, semicolons, or newlines are passed verbatim. Interpolating the task into the command line would let the shell reinterpret it.

What are the limits of the Aurora team orchestration engine?

The token budget is capped at 500K, overridable via environment variable, and runs take minutes so a long Bash timeout is required. It only runs your authenticated Claude Code and never substitutes third-party models or agents.