AgentSkillsCN

prime-cli

使用 Prime Intellect CLI 设置实验室工作空间、管理验证环境、运行评估，并启动托管的强化学习训练。

SKILL.md
--- frontmatter
name: prime-cli
description: Use the Prime Intellect CLI to set up Lab workspaces, manage verifiers environments, run evaluations, and launch hosted RL runs.

Prime CLI

Use this skill when the user asks about the Prime Intellect CLI, Lab, verifiers environments, evaluations, or hosted training.

Quick Flow

  1. •Read local docs: AGENTS.md and environments/AGENTS.md in the workspace.
  2. •Identify which action is needed: setup, environment init or install, evaluation, or RL run.
  3. •Run the minimal prime subcommand and report results concisely.

Common Commands

text
prime --version
prime login
prime lab setup

prime env init <env-name>
prime env install <env-name-or-hub-id>
prime env push --path ./environments/<env-name>

prime eval run <env-name-or-hub-id> -m <model>
prime eval tui

prime rl models
prime rl run <config.toml>

Workspace Conventions

  • •prime lab setup creates configs/, environments/, and top-level AGENTS.md.
  • •Environment modules expose load_environment and live under ./environments/<env-name>/.
  • •Evaluation outputs go under ./outputs or ./environments/<env-name>/outputs when configured.

Tips

  • •If the user mentions a specific model, verify availability with prime rl models.
  • •For custom endpoints, point to configs/endpoints.py.
  • •Keep responses short and list the exact commands used.

Examples

  • •'Initialize a new environment called math-probe' -> prime env init math-probe
  • •'Run an eval on my env using gpt-5-nano' -> prime eval run my-env -m gpt-5-nano
  • •'Start a hosted RL run' -> prime rl run configs/<run>.toml

References

  • •AGENTS.md (workspace root)
  • •environments/AGENTS.md