skills.rest
Lingjie-wang
Lingjie-wangCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

rl-env-integrator

Integrate RL environments behind a Gymnasium adapter contract with extracted specs and smoke tests.

Updated Jul 9, 2026
Search Tags:#reinforcement learning#adapter#smoke test#spec extraction#gymnasium#pettingzoo#environment integration
One-click install
npx skills add https://github.com/Lingjie-wang/autoRL --skill rl-env-integrator-lingjie-wang
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rl-env-integrator
Source: https://github.com/Lingjie-wang/autoRL/tree/main/skills/rl-env-integrator
Command: npx skills add https://github.com/Lingjie-wang/autoRL --skill rl-env-integrator-lingjie-wang

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Getting a concrete RL environment (official benchmark, custom code, or external simulator) into a constructible, spec-documented, smoke-tested state is error-prone: documentation disagrees with runtime behavior, dependency versions break, and specs get hand-written instead of measured. This Skill standardizes that integration behind a single adapter contract and reuses prior verified integrations when possible. ## Core Features & Use Cases - Verified Integration Reuse: Runs a deterministic reuse gate (scripts/find_reusable_integration.py) that links a previously verified integration into a new run only when artifacts, verification reports, metadata, and source hashes all match. - Adapter Contract Deliverables: Produces adapter.py (single make_env entrypoint), env_config.json with pinned dependencies, extract_spec.py that derives env_spec.json from the live environment, and smoke_rollout.py for random-policy checks. - Route Classification: Handles official_benchmark, custom_env, external_simulator, PettingZoo multi-agent, and EPyMARL channels, with a known-pitfalls knowledge base covering Python version pins, lazy registration, and reward-bound measurement. - Use Case: Given a custom thermal-control environment file plus a natural-language description, wrap it in the Gymnasium API, extract an honest spec from the live object, run a smoke rollout, and hand off a complete integration report for independent verification. ## Quick Start Use the rl-env-integrator skill to integrate the environment described in my task card into runs/<task-id>/, reusing a verified integration if one exists.

Frequently Asked Questions about rl-env-integrator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I integrate a custom RL environment with Gymnasium?▼

Subclass gymnasium.Env, declare observation_space and action_space from real data samples, and map the native lifecycle to reset returning (obs, info) and step returning the 5-tuple with terminated and truncated split. Expose everything through a single make_env(config) adapter entrypoint.

How to extract an environment spec from a Gymnasium environment?▼

Build the environment through the adapter and read spaces, reward bounds, and termination limits from the live object, never from documentation. Measure observed_reward_bounds over at least 10 full episodes and record fields like observation_modality, action_type, and randomness_sources.

Why does gym.make fail with NameNotFound for an installed package?▼

Packages like ale-py, minigrid, and gymnasium-robotics only register their environment ids after gymnasium.register_envs(<pkg>) runs. Call it lazily inside make_env behind a module-level flag, never at import time, so multiple adapters stay importable in one process.

Can I reuse a previously verified RL environment integration?▼

Yes, run the find_reusable_integration.py finder, which requires complete artifacts, a passing non-stale verification report, matching metadata and config constraints, and matching source hashes for custom environments. Eligible runs are linked read-only via symlink with an environment_reuse.json receipt.

Why is env.spec.max_episode_steps None for Atari or MiniGrid?▼

These environments truncate internally rather than via a TimeLimit wrapper: ALE uses max_num_frames_per_episode divided by frameskip, and MiniGrid uses env.unwrapped.max_steps. Spec extraction and smoke scripts must carry their own explicit step cap instead of looping unboundedly.

What are the limitations of environment integration before training?▼

Integration only makes the environment constructible, spec-documented, and smoke-tested; it does not choose algorithms, write training loops, or claim performance. A passing smoke rollout is not verification, and integration is not complete until an independent verifier passes.

Related Skills

Explore More Software Engineering Skills→
nabitllcnabitllc

todero-converting-plans-to-tasks

Convert Todero plans into executable issue graphs with dependencies and assignments.

Official
Intermediate
bfly123bfly123
3.5k

round-verification

Verifies a CCB execution round and produces a machine-readable round result artifact.

Community
Basic
simular-aisimular-ai
101

simulang

Run Simulang scripts with pinned simulang-js versions via CLI.

Official
Intermediate
LeonChaoXLeonChaoX
783

pufferlib

Accelerate reinforcement learning with parallel environment simulation and PPO training.

Community
Advanced