AgentJet

Execute AgentJet reinforcement learning experiments using experiment blueprints in classic (non-swarm) mode. Handles full lifecycle: launch experiment in tmux, monitor progress, analyze errors, collect results, and write finish flag. Use when the user wants to run AgentJet training experiments without the swarm distributed framework.

Execute AgentJet reinforcement learning experiments using experiment blueprints in swarm mode. Handles full lifecycle: generate blueprint if needed, launch experiment in tmux, monitor progress, analyze errors, collect results, and write finish flag. Use when the user wants to run or debug AgentJet training experiments.

Install AgentJet swarm server using Conda. Handles Python 3.10 environment creation, dependency installation with the verl training backbone, flash-attn compilation, and optional PyPI mirror for China users.

Install and run the AgentJet Swarm Server in a Docker container with NVIDIA GPU support. Use when the user wants to deploy a swarm server on a GPU machine via Docker, including GPU driver setup, Docker mirror configuration, model weight mounting, and server startup.

Download per-step time-series metric data (reward, entropy, response length, etc.) from a SwanLab cloud run URL as a pandas.DataFrame. Use when the user provides a SwanLab URL and wants to fetch or analyze training curves.

Install AgentJet client for connecting to a swarm server. Use when the user only needs to run the AgentJet client (not a swarm server) and does not need to run models locally, e.g. on a laptop. Installs basic requirements via `pip install -e .`.

Map VERL training configuration to AgentJet configuration. Find VERL config in verl_default.yaml, check for existing mappings in config_auto_convertion_verl.jsonc, add new mappings to ajet_default.yaml and the conversion schema, and optionally add parameters to AgentJetJob.

Convert skills in non-standard formats to the standard Agent Skills `SKILL.md` format. Validates YAML frontmatter (name, description, license, compatibility, metadata, allowed-tools), directory structure (SKILL.md, scripts/, references/, assets/), and best practices. Use when the user asks to normalize, validate, or fix a skill.

Train complex blackbox agents (agents without clear reward signals) using AgentJet. Write dataset collectors, episode runners with LLM-as-Judge reward functions, and integrate with the AgentJet training loop.

Install AgentJet swarm server using the UV package manager. Handles virtual environment creation with Python 3.10, dependency installation with the verl training backbone, flash-attn compilation, and optional PyPI mirror for China users.

Create an active, dataset-driven AgentJet swarm client. Write agent_roll.py and agent_run.py that iterate through a dataset, execute agent workflows, and compute rewards for reinforcement learning training with AgentJet Swarm.

How `max_env_worker` caps the "Running Episodes" gauge, and how `AgentJetJob` relates to the YAML config.

Monitor training progress by reading tmux content with exponential backoff intervals (30s, 1min, 2min, 4min, 8min, 16min), analyze logs when anomalies occur, and provide fix suggestions

Create a passive swarm client that waits for user input instead of iterating through a dataset by itself.