메뉴
Use as the top-level guide for ALab agent-facing role skills. It explains how to install the ALab CLI package, the root/project/experiment skill hierarchy, when to use each subskill, and how the three role skills differ without loading every command reference.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | alab-skills |
| description | Use as the top-level guide for ALab agent-facing role skills. It explains how to install the ALab CLI package, the root/project/experiment skill hierarchy, when to use each subskill, and how the three role skills differ without loading every command reference. |
ALab installs as the alab-cli Python package and exposes the alab console command.
Install it from PyPI:
python -m pip install alab-cli
alab help
ALab is a local agent-first experiment workbench. It separates root home administration, project coordination, and experiment worktree work so agents can collaborate without sharing unnecessary credentials or mutating the wrong layer.
Use this top-level skill when you need to understand the ALab role-skill system before choosing a specific subskill. For actual operations, load the narrowest matching subskill and its command reference only when needed.
alab-global-admin: Root layer. Use for ALab home bootstrap, root credential rotation, project admin key creation/revocation, project initialization, SkyDiscover catalog lifecycle, global cache/backup pruning, root dashboard, and root/global audit inspection. It should hand project coordination to alab-project-controller and worktree work to alab-experiment-worker.alab-project-controller: Project layer. Use with one project admin key to create and coordinate experiments, manage project-scoped config/source/validation/lifecycle state, observe evidence, compare runs, and launch experiment worker sessions or subagents. It must not perform root administration or do experiment work in the current session unless the user explicitly asks.alab-experiment-worker: Experiment worktree layer. Use inside one experiment worktree with that worktree token context to inspect visible evidence, edit candidate source, run evaluation, annotate useful context, and submit final results. It must not accept project admin/root keys or perform project/root operations.alab-global-admin; project admin key implies alab-project-controller; experiment worktree token context implies alab-experiment-worker.SKILL.md: This top-level ALab skill guide and subskill router.SKILL_cn.md: Synchronized Chinese version of this file.agents/openai.yaml: UI metadata and default prompt for this top-level skill.alab-global-admin/: Root-layer subskill, including its SKILL.md, SKILL_cn.md, agents/openai.yaml, and references/commands*.md.alab-project-controller/: Project-layer subskill, including its SKILL.md, SKILL_cn.md, agents/openai.yaml, and references/commands*.md.alab-experiment-worker/: Experiment-worktree subskill, including its SKILL.md, SKILL_cn.md, agents/openai.yaml, and references/commands*.md.Use when administering an ALab home with root authority, including home bootstrap, root and project-admin credential management, project initialization and handoff, SkyDiscover catalog management, global cache or backup pruning, and root-level audit/dashboard inspection. Do not use for routine project coordination or experiment work.
Use when managing one existing ALab project with a project admin key to create and coordinate experiments, validate and adjust project configuration, manage project-scoped source or lifecycle state, observe project evidence, and launch experiment worker sessions or subagents without exposing admin credentials.
Use when operating inside one ALab experiment worktree with that worktree token context to inspect status, edit candidate source, run evaluations, submit final results, and read visible experiment evidence without project admin or root privileges.