team-overnight

Orchestrates a persistent multi-agent team for unattended overnight Astroray development runs.

Updated Dec 27, 2025
One-click install
npx skills add https://github.com/HendrikGC02/Astroray --skill team-overnight-hendrikgc02
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: team-overnight
Source: https://github.com/HendrikGC02/Astroray/tree/main/.claude/skills/team-overnight
Command: npx skills add https://github.com/HendrikGC02/Astroray --skill team-overnight-hendrikgc02

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating unattended multi-hour development work on the Astroray renderer requires dispatching packages, verifying GPU renders, merging PRs, and writing status reports without human supervision, which a single agent session cannot safely do alone. ## Core Features & Use Cases - Persistent agent team: Spawns named long-lived teammates (architect, pr-merger, docs-scribe) addressable via SendMessage, plus ephemeral implementers and hardware-verifiers per tick. - Hardware-safe GPU serialization: Enforces a single CUDA job at a time through the orchestrator GPU lock, preventing false-positive illegal-access crashes on the RTX card. - Gated autonomous merging: Auto-merges PRs only when CI passes, hardware verification is bound to the head SHA, and independent review signs off; HW failures are never overridden. - Use Case: Kick off the skill before leaving for the night; the architect researches and specs the next packages, the orchestrator dispatches implementations in isolated worktrees, hardware-verifier visually checks rendered PNGs, and docs-scribe delivers a morning standup of shipped and blocked items. ## Quick Start Run /team-overnight after confirming a clean main branch, free orchestrator locks, a disabled Astroray-RoadmapOrchestrator scheduled task, and a recently built .pyd.

Frequently Asked Questions about team-overnight

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

FAQPage Schema
How do I run an unattended multi-agent coding session overnight?▼

Invoke /team-overnight after verifying preconditions: clean main branch, free orchestrator and GPU locks, no leftover worktrees, and a recent .pyd build. The team-lead then drives the orchestrator loop on a 10-minute cadence until last-call.

How do I spawn persistent named agents that survive task completion?▼

Spawn each teammate with the Agent tool using the name parameter and run_in_background; the name makes the agent addressable via SendMessage, which resumes it from its transcript. Do not call TeamCreate or pass team_name, as both are deprecated.

Can two GPU render verification jobs run in parallel?▼

No. Two concurrent CUDA-heavy verifiers on this RTX card cause false-positive illegal-access crashes, so every hardware job must acquire the orchestrator GPU lock first. The lock is non-negotiable for all team members.

What conditions must a PR meet before auto-merging?▼

Auto-merge requires all CI checks passing, a hardware PASS result bound to the PR head SHA, and for non-HW-gated PRs an independent different-model sign-off. A hardware FAIL is never overridden automatically.

What happens if the GPU lock gets stuck during a run?▼

A GPU lock held over 90 minutes indicates the verifier crashed. Inspect the lock metadata file, release it manually, and the next orchestrator tick re-evaluates from ledger state without corrupting work.