GitHub project orchestrates Codex with Astra as planner and Luna subagents
TheMoonMidas · x · 2026-09-07
Developer melvindvivas updated the open-source project codex-astra-luna-orchestrator (229 stars): a Codex setup where GPT-6 Astra acts as root orchestrator and reviewer while GPT-5.6 Luna is pinned as the default model for execution subagents.
- Configured via .codex/config.toml: main model gpt-6-astra (reasoning effort low), agents enabled, up to 4 concurrent subagent threads per session, subagents default to gpt-5.6-luna (medium).
- Each role (explorer/worker/tester/reviewer/researcher) has its own toml pinning the intended model: Luna handles exploration, execution, testing and research; Astra handles review.
- Ships with an astra-orchestrator skill, AGENTS.md, and setup.sh/setup.ps1 installers. The update enforces Luna usage more strictly for research and exploration tasks.
More from coding & agent
- Dev Builds a Secret Agent Message Board Inside His Personal Linear Team — willcb · 2026-09-07
- Free 155-page Vibe Coding handbook covers specs, tests, deploy: idea to production in 15 chapters — Mahmoud_Zalt · 2026-09-07
- Agent-written Stripe integration caused duplicate charges; sandbox twins fix it — Common_Dream9420 · 2026-09-07
- Qwen Code ships v0.23.1-preview.2 with live subagent status and workflow visualization — qwen-code-ci-bot · 2026-09-07
- The AI engineering methodology ladder: 92% of US devs now vibe coding — No_Stretch433 · 2026-09-07
- APT-RAG Builds Adaptive Reasoning Trees for QA Over Hundreds of Documents — _reachsumit · 2026-09-07