# Add Example Doc Model Env

> Adds example documentation for a new model or environment in RLinf (RST pages in the docs gallery for both English and Chinese). Use when adding a new embodied or reasoning example, or new benchmark (e.g. LIBERO, ManiSkill).

- Skill: `rlinf/add-example-doc-model-env` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add rlinf/add-example-doc-model-env`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rlinf/add-example-doc-model-env/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: rlinf (https://skillmd.com/u/rlinf)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rlinf/add-example-doc-model-env

---


# Add Example Doc to a Model or Environment

Use this skill when adding example documentation for a new **model** (e.g. π₀, GR00T, OpenVLA) or **environment** (e.g. LIBERO, ManiSkill, MetaWorld) in RLinf. Documentation is added for both **English** and **Chinese**.

Before writing, read `docs/STYLE_GUIDE.md` and apply the `refine-docs` natural-language
gate. Explain the reader's task before introducing implementation names; do not
put an unexplained API term in a heading. Write English and Chinese independently
from the same meaning instead of translating clause by clause. Keep familiar
developer terms in English when an unusual translation would obscure the code.

---

## Steps

1. **Create the English RST file**  
   Examples are now grouped by **category**:
   - `embodied/` – embodied RL/VLA examples (e.g. ManiSkill, LIBERO, Dexbotic, π₀, OpenSora)
   - `agentic/` – agent / tool-use / coder / math reasoning examples (e.g. SearchR1, coding_online_rl, reasoning)
   - `system/` – placement, scheduling, system demos  
   Path pattern: `docs/source-en/rst_source/examples/<category>/<name>.rst`  
   - Example (embodied): `docs/source-en/rst_source/examples/embodied/dexbotic.rst`  
   - Example (agentic): `docs/source-en/rst_source/examples/agentic/searchr1.rst`  
   Follow `docs/STYLE_GUIDE.md` first, then use current examples in the same category
   for category-specific details (see [reference.md](reference.md)).

2. **Register in the English category index**  
   The embodied gallery is split into five category index files that live directly under `examples/` (not inside `embodied/`):
   `simulators_index.rst` (benchmark/simulator-centric), `real_world_index.rst` (real-robot hardware),
   `vla_wam_index.rst` (model-centric: π₀, GR00T, …), `sft_index.rst` (SFT recipes), and
   `methods_index.rst` (algorithm-centric: DAgger, RECAP, IQL, …). The example RST itself still lives at
   `docs/source-en/rst_source/examples/embodied/<name>.rst`; the index files reference it as `embodied/<name>`.
   Non-embodied categories (agentic, system) still use `docs/source-en/rst_source/examples/<category>/index.rst`.
   - Edit the matching index file (e.g. `docs/source-en/rst_source/examples/vla_wam_index.rst` for a new VLA model, or `examples/<category>/index.rst` for agentic/system).
   - Add an entry for `<name>` in the hidden `.. toctree::` at the bottom (for embodied, prefixed: e.g. `embodied/dexbotic`).
   - Optionally add a gallery card in the correct section using the same HTML block pattern as existing cards (image, hyperlink to the rendered doc — `embodied/<name>.html` for embodied — short title + description).  
   **Note:** The top-level `examples/index.rst` only links to the category indexes and usually does not need to be changed when adding a single example.

3. **If it is an embodied evaluation environment**  
   In `docs/source-en/rst_source/evaluations/index.rst` or the relevant evaluation guide, add the new evaluation flow:
   - For embodied examples: `:doc:\`Display Name <../examples/embodied/<name>\``
   - For other categories, follow the existing pattern in that file and mirror the relative path used there.

4. **Create the Chinese RST file**  
   Use the same `<category>` and `<name>` as in English:  
   Path: `docs/source-zh/rst_source/examples/<category>/<name>.rst`.  
   Keep the same structure, sections, commands, and technical meaning as English,
   but rewrite the prose in natural Chinese. Do not mirror English sentence order.
   Use existing EN/ZH pairs under the same category (e.g. `embodied/libero.rst` in
   both `source-en` and `source-zh`) as reference.

5. **Register in the Chinese category index**  
   Edit the matching Chinese index file — for embodied, the same five split indexes under `docs/source-zh/rst_source/examples/` (`simulators_index.rst`, `real_world_index.rst`, `vla_wam_index.rst`, `sft_index.rst`, `methods_index.rst`); for agentic/system, `docs/source-zh/rst_source/examples/<category>/index.rst`:  
   - Add the same `<name>` entry to the hidden `.. toctree::`.  
   - If you added a gallery card in the English index, add a matching Chinese gallery card here, following existing HTML patterns.  
   If the Chinese evaluation docs list the same benchmark or environment, add the new example using the same relative path pattern as in English.

6. **Update README.md**  
   In the "What's NEW!" section at the top, add a new dated bullet, with the documentation link pointing to the correct **category** path, e.g.:  
   - `- [YYYY/MM] 🔥 ... Doc: [Display Title](https://rlinf.readthedocs.io/en/latest/rst_source/examples/embodied/<name>.html).`  
   If the example is a simulator, model, or feature that appears in the Key Features table, add a corresponding list item in the right column using the same category path (see [reference.md](reference.md)).

7. **Update README.zh-CN.md**  
   In the "最新动态" section, add the same news item in Chinese with the doc link using `/zh-cn/` and the correct category path, for example:  
   - `https://rlinf.readthedocs.io/zh-cn/latest/rst_source/examples/embodied/<name>.html`  
   If you added a feature list entry in README.md, add the same entry in README.zh-CN.md (Chinese display text, `zh-cn` in the link, and the same `<category>` segment).

### RST structure (concise)

- Title (overbar length matches title).
- Optional HuggingFace icon block (copy from libero.rst).
- Short intro (what this example does, which model + env).
- **Environment**: env name, task, observation/action space, task description format, data shapes.
- **Algorithm**: PPO/GRPO/etc. and model architecture notes.
- **Dependency Installation**: clone, install (Docker or pip).
- **Quick Start**: exact commands and key YAML/config snippets.
- **Evaluation** (if applicable): eval command and config notes.

Use existing examples in the same category (e.g. `embodied/libero.rst`, `embodied/pi0.rst`, `agentic/searchr1.rst`) as templates; see [reference.md](reference.md) for a minimal template.

---

## Checklist

- [ ] English RST created: `docs/source-en/rst_source/examples/<category>/<name>.rst`.
- [ ] English category index updated: `docs/source-en/rst_source/examples/<category>/index.rst` (toctree; optional gallery card).
- [ ] If embodied eval env: `docs/source-en/rst_source/evaluations/index.rst` or the relevant evaluation guide updated with the new flow.
- [ ] Chinese RST created: `docs/source-zh/rst_source/examples/<category>/<name>.rst`.
- [ ] English and Chinese explain concepts before naming APIs; headings contain no unexplained implementation terms.
- [ ] Chinese reads naturally and retains familiar, searchable English developer terms.
- [ ] Chinese category index updated: `docs/source-zh/rst_source/examples/<category>/index.rst` (toctree; gallery card if added for EN).
- [ ] If embodied eval env: Chinese evaluation page updated if it lists environments (use the same relative path pattern as EN).
- [ ] README.md updated: new bullet in "What's NEW!" and, if applicable, entry in Key Features table (using the correct category path).
- [ ] README.zh-CN.md updated: new bullet in "最新动态" and, if applicable, entry in 核心特性 table (using the correct category path).