docs(agents): record W&B product and platform facts for style passes

AGENTS.md

@@ -85,7 +85,16 @@ To check or apply style, read the relevant pass file(s) and use them as guidance
 - **Python SDK functions**: Module-level functions in the Python SDK are listed in the [Global Functions overview](/models/ref/python/functions).
 - **`.editorconfig`**: An `.editorconfig` file in the repository root enforces indentation and whitespace automatically. Most editors apply it with no configuration. If yours doesn't support it natively, install the EditorConfig plugin (https://editorconfig.org/#download).
 - **Consistent language tab labels**: When an example offers multiple languages — in a `<CodeGroup>` or across `<Tab title="...">` blocks — label every tab with the same canonical name everywhere: `Python`, `TypeScript` (never `Typescript`), `Bash`. In a `<CodeGroup>`, give each fence a lowercase lexer **and** that canonical title (e.g. a `python` fence titled `Python`); never leave a language fence untitled. The reader's **Python/TypeScript** choice carries from page to page via `code-group-language-persist.js` (repo root), which matches those two labels case-insensitively — so inconsistent casing of them silently resets it. Other labels like `Bash` are only for in-page consistency and are intentionally not persisted across pages. Don't add a competing per-page persistence script.
-- **ARIA chat examples**: When a task can be delegated to ARIA end-to-end in the W&B app, add a compact chat example as a third content modality alongside code examples and UI click sequences. Use **ARIA** as the public-facing product name in prose and in fence titles.
+- **Company name**: W&B (not "Weights & Biases") in running text.
+- **First mention pattern**: Use "W&B [Product]" on first mention of a product in a page, then drop "W&B" for later mentions — this is how Weave, Models, and Launch behave. Run, artifact, and report are looser: keeping "W&B" throughout (not just on first mention) is current usage too. Sweep is the exception that always drops it — see Sweeps below.
+- **Product name capitalization**: Products stay capitalized on their own — W&B Weave → Weave, W&B Models → Models, W&B Launch → Launch. W&B artifact and W&B run are always lowercase ("an artifact", "a run"), matching how the API names these objects.
+- **The web app is "W&B"** ("Weights & Biases" in full) — not the "W&B App". Now that a mobile app exists, "app" no longer identifies the web product. Name the interface itself the **W&B UI**, a component of W&B, when a sentence needs to distinguish it from the API or SDK.
+- **W&B Python SDK**: The `wandb` package is the W&B Python SDK. Use "the W&B Python SDK (`wandb`)" as the full form and "the `wandb` library" as the short form.
+- **Weave UI labels**: When one of these words names a tab, view, page, table, or panel in the Weave UI, it keeps its UI casing: Traces, Calls, Threads, Conversations, Spans, Signals, Agents, Scorers, Monitors, Evals, Models, Datasets, Prompts, Assets, Dashboard, Playground, Leaderboard. The same words as ordinary nouns in prose ("the calls in this trace") are lowercase throughout the docset — don't recapitalize them.
+- **Project navigation**: The navigation inside a W&B project is the **project sidebar** (the **Weave project sidebar** in Weave docs), not the "left sidebar", "sidebar menu", or "left navigation". Bare "sidebar" is fine. This covers W&B's own in-project navigation only — leave third-party console steps (for example, the Microsoft Entra ID and Slack procedures under `platform/hosting/`), W&B organization- and team-level navigation, and `release-notes/` wording as written.
+- **Sweeps**: W&B Sweeps is the product; an individual one is "a sweep", not "a W&B sweep". This is specific to sweeps — "a W&B run", "a W&B artifact", and "a W&B report" are current usage.
+- **Generated content**, to fix upstream rather than here: the directories in README.md's [What files do I edit?](README.md#what-files-do-i-edit) table, `snippets/_includes/code-examples/` and `snippets/CodeSnippet.jsx` (synced from `wandb/docs-code-eval`), and any table wrapped in a `{/* takeru … automatically generated … */}` marker under `inference/` and `serverless-training/`. Pages in `weave/cookbooks/` are notebook-derived: each has a Jupyter notebook source at `weave/cookbooks/source/<page-name>.ipynb`.
+- **ARIA chat examples**: When a task can be delegated to ARIA end-to-end in the W&B UI, add a compact chat example as a third content modality alongside code examples and UI click sequences. Use **ARIA** as the public-facing product name in prose and in fence titles.
 
   Default format: a user prompt and a concise ARIA response. Don't include reasoning or thinking steps unless the page specifically needs them for clarity; longer walkthroughs belong on [ARIA overview](/aria/overview). Place chat examples in the first section where the ARIA-delegable task appears, not at the top of the page unless the whole page is about chatting with ARIA.
 
@@ -113,7 +122,7 @@ To check or apply style, read the relevant pass file(s) and use them as guidance
 
   - **Regular reports only, not Fully Connected articles** — FC articles keep their full blog chrome in a frame and look broken.
   - **Anonymous-viewable** — a public-project report or a [view-only link](/models/reports/cross-project-reports#view-only-report-links). The URL (with any `?accessToken=`) ships in public source and git history, so treat the report as public forever.
-  - **One or two per page**, skinny and purpose-built — each iframe boots the full W&B app at a fixed height, in its own light theme.
+  - **One or two per page**, skinny and purpose-built — each iframe boots the full W&B UI at a fixed height, in its own light theme.
   - **English sources only** — not `ja/`, `ko/`, `fr/`.
 
   A weekly CI job (`scripts/report-embeds/check_embeds.py`) verifies each embedded report still renders and files an issue if one breaks. It does not gate PRs.