AGENT SKILL

cartpole-classic-control

Operate DQN, A2C, and PPO CartPole-v1 training, testing, checkpoints, and smoke diagnostics for the reinforcement-learning repo.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
cartpole-classic-control
license
MIT
metadata
{ "disco-role": "operating" }
description
Operate DQN, A2C, and PPO CartPole-v1 training, testing, checkpoints, and smoke diagnostics for the reinforcement-learning repo.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

CartPole Classic Control

Use this sub-skill when the task involves the repository's CartPole-v1 DQN, A2C, or PPO workflows: choosing the right algorithm script, explaining network/update behavior, running training or replay, diagnosing checkpoints, or doing a safe headless smoke check.

Do not use it for GridWorld dynamic programming or tabular control, standard Atari Breakout/Pong CNN workflows, hard-exploration Atari/RND/Go-Explore/robustification, or paper-level benchmark claims.

Fast route

Operating guardrails

  • Treat full CartPole training as a real run, not a smoke test. DQN can run up to 300 episodes, A2C up to 1000, and PPO up to 1500 update cycles with 1024 environment steps per update.
  • Avoid --render during training unless the user explicitly wants an interactive window; it slows learning and can fail on headless hosts.
  • --test also requests Gymnasium render_mode="human"; use it only when a display is available and a compatible checkpoint is present.
  • The bundled smoke script is intentionally standalone and does not require Gymnasium, Pygame, the training scripts, or a display. It proves tensor/API shape and update invariants, not training convergence.
  • Checkpoints are algorithm-specific: DQN and PPO save raw model state_dicts; A2C saves a dictionary with separate actor and critic entries.

Provenance labels

This sub-skill distills the CartPole workflows identified in the source material as 2-cartpole/1-dqn.py, 2-cartpole/2-a2c.py, 2-cartpole/3-ppo.py, and the shared CartPole environment helper 2-cartpole/env.py. Those names are provenance/workflow labels; the runtime guidance below is self-contained and does not require opening the original source files.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351