---

name: environment-api description: "Use Gymnasium's single-environment API: make/reset/step/render/close loops, custom Env classes, registration/specs, checker validation, seeding, and old Gym migration." disable-model-invocation: true metadata: disco-role: operating license: MIT

Gymnasium Environment API

Use this sub-skill when a task involves creating or interacting with a single Gymnasium Env, migrating old Gym code, registering a custom environment, inspecting EnvSpec, or fixing reset/step/check_env failures.

Start Here

• For a basic loop, use gymnasium.make(...), obs, info = env.reset(...), and obs, reward, terminated, truncated, info = env.step(action); see references/api-reference.md.
• For custom environments, subclass gymnasium.Env, define action_space and observation_space, call super().reset(seed=seed), and validate with check_env; see references/custom-environments.md.
• For migration from old Gym, update imports, reset return unpacking, seeding, render mode placement, and the done split; see references/migration-and-validation.md.
• For common errors and warnings, use references/troubleshooting.md before changing unrelated code.
• To smoke-test a minimal custom env from this sub-skill directory, run python scripts/check_custom_env.py --help and then python scripts/check_custom_env.py.

Scope Boundaries

• Use ../spaces-data/SKILL.md for detailed space constructors, flattening, JSON conversion, dtype/shape design, and contains diagnostics.
• Use ../wrappers-recording/SKILL.md for wrapper catalogs, wrapper subclassing depth, recording videos, rendering wrappers, and media dependencies.
• Use ../vectorization/SKILL.md for gym.make_vec, SyncVectorEnv, AsyncVectorEnv, batched actions/observations, and vector wrappers.
• Use ../builtin-envs/SKILL.md for choosing built-in environment families, optional extras, Atari/ALE, Box2D, MuJoCo, Toy Text masks, and plugin install issues.

Minimum Correct Patterns

import gymnasium as gym

env = gym.make("CartPole-v1")
obs, info = env.reset(seed=123)
terminated = truncated = False
while not (terminated or truncated):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
env.close()

from gymnasium.utils.env_checker import check_env

check_env(raw_env, skip_render_check=True)

Keep public runtime guidance self-contained. Do not depend on the Gymnasium source tree, test tree, or documentation paths at runtime.