Runs a four-pass claude -p chain that frames, explores, red-teams, and synthesizes one topic.
npx skills add https://github.com/Kaleb-Rupe/aurora --skill deep-kaleb-rupe
Please help me install this Agent Skill. Skill: deep Source: https://github.com/Kaleb-Rupe/aurora/tree/main/claude/skills/deep Command: npx skills add https://github.com/Kaleb-Rupe/aurora --skill deep-kaleb-rupe
aurora CLI using only your authenticated Claude Code (claude -p), with no third-party models, no UA spoofing, and a budget cap of 300K tokens (env-overridable). - Safe Input Handling: Passes the topic through a quoted heredoc on stdin so quotes, backticks, and shell metacharacters in the user's text never reach the shell. - Use Case: Ask to deeply analyze whether a microservices migration is justified for your team, and receive a live 4-pass tree followed by a synthesis that has already survived a red-team pass. ## Quick Start Ask the assistant to go deep on a single topic, for example by typing /deep followed by the question you want pressure-tested.High-intent search queries and answers about installing and using this skill.
Type /deep followed by your topic, or ask to deeply analyze one question. The Skill invokes the aurora CLI in deep mode, passing your topic verbatim through a quoted heredoc on stdin, and relays the live 4-pass tree plus final synthesis.
Pass 0 frames the question, pass 1 explores it, pass 2 red-teams the exploration, and pass 3 synthesizes an answer that withstands the critique. Each pass resumes the prior session so reasoning builds cumulatively.
No, it uses only your authenticated Claude Code via claude -p calls. There is no third-party model or agent, no user-agent spoofing, and the token budget is capped at 300K, overridable via environment variable.
The quoted 'AURORA_TASK' heredoc delimiter disables all shell expansion, so quotes, dollar signs, backticks, semicolons, and newlines in your topic cannot reach the shell. This prevents command injection from arbitrary user text.
A real run takes minutes because it executes four sequential claude -p passes. Use a long Bash timeout when invoking it, and expect streaming stdout showing the live pass tree before the final synthesis.
Nikhil-Chopra-7Sanitize meeting transcripts and extract structured decisions, actions, and feedback.
ryan-mahoneyAnalyze architecture proposals from two practitioner perspectives and produce a structured critique document.
plurigridApply System2-attention to regenerate context and refine transformer attention weights.
jonnytacos8Run non-interactive Q&A, summarization, and generation via Gemini CLI.