"Routes DI-engine users to the right workflow for CLI/config
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill di-engine --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Di Engine?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-di-engine)More formats (shields.io, HTML) on the badges page.
---
name: di-engine
description: "Routes DI-engine users to the right workflow for CLI/config
launches, serial and framework-based RL pipelines, and environment
integration."
metadata:
disco-role: operating
disable-model-invocation: true
license: Apache 2.0
disable-model-invocation: true
---
# DI-engine
Use this skill for the public DI-engine reinforcement-learning engine.
It routes common user requests to focused sub-skills instead of forcing future
agents to rediscover the repo structure from scratch.
## Quick check
If you only need to verify that the package is installed, start with:
```bash
python -m pip check
python -c "import ding, dizoo; print(ding.__version__)"
```
For a faster bundled smoke check, run `scripts/check_install.py` from an
environment where DI-engine is installed.
## Main routes
### `cli-config`
Read this when the user asks about:
- `ding` or `ditask` CLI usage
- config files, `main_config`, `create_config`, or `system_config`
- `compile_config`, `read_config`, `save_config`, or registry queries
- launch flags for `serial`, `parallel`, `dist`, `eval`, or `ditask`
- the repo-provided shell wrappers for serial/parallel launches
### `serial-pipelines`
Read this when the user asks about:
- `ding.entry.serial_pipeline`, `serial_pipeline_onpolicy`, or `serial_pipeline_offline`
- `eval`, `collect_demo_data`, `collect_episodic_demo_data`, or episodic post-processing
- legacy `ding/entry/*` launchers and environment-specific entry scripts under
`dizoo/`
- special modes such as SQIL, GAIL, DQFD, TREX, NGU, reward-model, or offline RL
### `framework-runtime`
Read this when the user asks about:
- `ding.framework.task`, `Parallel`, `Supervisor`, or `EventLoop`
- middleware-based recipes using `StepCollector`, `OffPolicyLearner`,
`multistep_trainer`, `interaction_evaluator`, `CkptSaver`, or `data_pusher`
- the modern example scripts in `ding/example/`
- multi-process message routing, `task.wait_for`, `task.emit`, or auto-recovery
### `env-integration`
Read this when the user asks about:
- `DingEnvWrapper`, `BaseEnvManager`, `BaseEnvManagerV2`, or `EnvSupervisor`
- env wrapper selection, reset/step shape mismatches, or new environment onboarding
- DIZoo environment patterns in `dizoo/classic_control/`, `bitflip/`,
`frozen_lake/`, or `league_demo/`
- custom environment adapters and the env-specific config/template pattern
## Useful bundled helpers
- `scripts/check_install.py` — quick import/version/config smoke for a prepared environment.
- `scripts/run_serial.sh` — thin wrapper for `ding -m serial` with a config path and seed.
- `scripts/run_parallel.sh` — thin wrapper for `ding -m parallel` with a config path and seed.
## Read first
- `references/package-surface.md` for the repo module map and public entry points.
- `references/troubleshooting.md` for cross-cutting install/import/runtime failures.
## Notes
- This skill covers the core DI-engine workflows that can be exercised from the
installed package and bundled helpers without reopening the source checkout.
- Advanced external environment families that need separate optional packages
are intentionally left out of the default routes.
- If a request mentions a specific example, config, or entry script, route to the
sub-skill that owns that workflow rather than reading the source tree directly.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!