feat(nodes)!: retire 24 nodes the sandbox and Code node already cover

docs/CLOUD_NODE_CURATION.md

@@ -148,7 +148,7 @@ that egress policy is written down in the ComfyUI row of
 | ----------- | ----------------------------------------------------- |
 | `lib.image` | Photoshop-style ops: warp, color, draw, effects, …    |
 | `lib.svg`   | SVG / vector graphics                                  |
-| `lib.grid`  | Image grid combine/slice                               |
+| `lib.grid`  | Image grid slicing                                     |
 | `lib.audio` | Audio DSP/effects (reverb, delay, EQ, …)               |
 
 ### Provider node namespaces
@@ -161,12 +161,11 @@ that egress policy is written down in the ComfyUI row of
   `lib.markdown`, `lib.html`, `lib.charts`
 - **System/automation:** `lib.os`, `nodetool.workspace`,
   `nodetool.triggers`, `lib.browser`, `lib.video.download`
-- **Databases/cloud/integrations:** `lib.sqlite`, `lib.http`, `lib.graphql`,
-  `lib.mail`, `lib.secret` (`lib.comfy` is dropped as a namespace, but its two
+- **Databases/cloud/integrations:** `lib.sqlite`, `lib.http`, `lib.graphql`
+  (`lib.comfy` is dropped as a namespace, but its two
   runners are allowlisted by name — see above)
 - **Messaging:** `messaging.discord`, `messaging.telegram`
-- **NLP/ML utility:** `lib.nlp`, `vector` (RAG),
-  `lib.validate`, `lib.datetime`
+- **NLP/ML utility:** `vector` (RAG), `lib.validate`, `lib.datetime`
 - **Out-of-scope providers:** `huggingface`, `transformers`, `minimax`, `reve`,
   `elevenlabs`, `replicate`, `together`, `topaz`, `atlascloud`
 

docs/CODE_NODE_COVERAGE.md

@@ -7,7 +7,10 @@ already have a snippet standing in for them?**
 
 **Status:** `nodetool.list` (`Range`, `Tile`, `RepeatEach`, `RepeatValue`),
 `lib.datetime` (all 5), and `lib.validate` (all 5) have been removed — their
-Tier 1/Tier 2 entries below are historical. `lib.rss` (`FetchRSSFeed`,
+Tier 1/Tier 2 entries below are historical. A later round removed the 12
+`lib.svg` element builders (step 2 of the order below), plus `lib.mail` (3),
+`lib.nlp` (7), `lib.secret.GetSecret`, and `lib.grid.CombineImageGrid` — see
+**Removed in the sandbox-coverage round** at the end. `lib.rss` (`FetchRSSFeed`,
 `ExtractFeedMetadata`) and the 27 `nodetool.data.*` dataframe verbs were also
 removed, on the same rationale but outside this audit's original scope: the
 sandbox's `@nodetool-ai/sandbox-xml` and `@nodetool-ai/sandbox-csv` packs
@@ -300,10 +303,11 @@ now covers this ground from inside a Code node. `ForEachRow` and
 | --- | --- |
 | `nodetool.constant.*` (except `Date`/`DateTime`), `nodetool.input.*`, `nodetool.output.*` | value editors, dropzones, pickers — pure UI |
 | `nodetool.control.*` (22) | actor-model stream semantics: fan-out, per-item emission, back-pressure. The sandbox runs once and returns once. The five Streaming snippets cover the generator patterns one node can express; the rest is kernel-level. |
-| `nodetool.image.*` `nodetool.audio.*` `nodetool.video.*` `nodetool.model3d.*` `lib.image.*` `lib.audio.*` `lib.grid.*` `nodetool.sketch/timeline/script` | content cards and bespoke editors; sharp/canvas/ffmpeg |
-| `lib.nlp.*` (7) | compromise, AFINN, stemmers, TF-IDF — real libraries |
+| `nodetool.image.*` `nodetool.audio.*` `nodetool.video.*` `nodetool.model3d.*` `lib.image.*` `lib.audio.*` `lib.grid.SliceImageGrid` `nodetool.sketch/timeline/script` | content cards and bespoke editors; sharp/canvas/ffmpeg. `lib.grid.CombineImageGrid` was **removed** — `image.grid` is the same sharp call, host-side. Snippet: `image-grid`. |
+| ~~`lib.nlp.*` (7)~~ | **Removed** without a replacement pack — the one open capability gap this round left. See below. |
 | `lib.pdf` `lib.charts` | PDF text extraction and page rasterization, and a chart renderer |
-| `lib.mail` `lib.apple` `messaging.*` | IMAP/SMTP, AppleScript, and long-lived bot connections — none of them a `fetch` call |
+| `lib.apple` `messaging.*` | AppleScript and long-lived bot connections — neither of them a `fetch` call |
+| ~~`lib.mail` (3)~~ | **Removed.** The argument below — "IMAP is not a `fetch` call" — was answered by something other than a guest script: `search_email` / `add_label_to_email` / `archive_email` already shipped as capabilities over the same `imapflow` client, and Gmail itself reaches a Code node through `@nodetool-ai/sandbox-nodetool/google`, where the OAuth token stays host-side. Snippets: `gmail-search`, `gmail-add-label`, `gmail-archive`. |
 | `lib.google` | **Removed.** Drive, Gmail, Docs, Sheets and Calendar are the `google` capability module: `import { drive_search } from "@nodetool-ai/sandbox-nodetool/google"`. The OAuth session stays host-side — the guest never holds the token. |
 | `lib.s3` `lib.supabase` `lib.notion` `lib.twilio` `messaging.*.SendMessage` `lib.mail.SendEmail` | **Removed.** Each was one authenticated HTTP call, so each is a Code node now: `fetch`, `nodetool.secrets.get(name)`, and the auth-helper packs (`@nodetool-ai/sandbox-aws` for SigV4, `-notion`, `-supabase`, `-twilio`). `lib.mail.SendEmail` went with them and has no guest path — SMTP is not HTTP; a script sends mail through an HTTP email API. |
 | `apify.*` `search.*` | **Removed.** Apify and SerpAPI are capability modules: `@nodetool-ai/sandbox-nodetool/apify` and `.../serpapi`. The token stays host-side — the guest never holds it. |
@@ -330,17 +334,20 @@ The six new categories are authored, so what remains is deletion and migration.
 1. **Tier 1 deletions** (39). No new snippets needed — the win is dropping the
    duplicate implementation, and `nodetool.text` is the namespace users hit
    first.
-2. **`lib.svg` deletion** (12). The cleanest of the new categories: one output
-   type, byte-identical markup, and the two rendering nodes stay.
+2. ~~**`lib.svg` deletion** (12).~~ **Done.** `Document` and `SVGToImage` stay;
+   the three shipped SVG examples now build their element list in one Code node,
+   where the array order is the paint order.
 3. **Tier 2 deletions** (39 more) — Path, Markdown, HTML, Validation and the
    singles, once each has a graph migration.
 4. **`workspace.copy`/`workspace.move`, then `lib.os` file ops** (16). Two
    bridge functions unlock the rest of the namespace.
 5. **Per-snippet output typing, then `nodetool.data`** (20). The one that needs
    design, not just authoring.
 
-`lib.http`, `lib.graphql` and `lib.secret.GetSecret` are deliberately absent
-from this list — their snippets shipped, their nodes stay.
+`lib.http` and `lib.graphql` are deliberately absent from this list — their
+snippets shipped, their nodes stay. `lib.secret.GetSecret` was in that sentence
+too, on the "its value is the picker rather than the call" argument; it has
+since been removed anyway — see below.
 
 ## What a removal needs alongside it
 
@@ -369,3 +376,44 @@ and the change reads as a regression.
   other half: describing a result instead of picking a snippet.
 - [CLOUD_NODE_CURATION.md](CLOUD_NODE_CURATION.md) — the cloud profile trims by
   namespace and admits the Code node by name.
+
+## Removed in the sandbox-coverage round
+
+Twenty-four node classes, ~5,300 lines. Each was covered by something already
+shipped, not by something this round had to build — except `lib.nlp`, which is
+recorded as a gap rather than a win.
+
+| Removed | Count | What covers it now |
+| --- | ---: | --- |
+| `lib.svg` element builders | 12 | The 12 `svg-*` snippets. `Document` and `SVGToImage` stay — the guest has no `sharp`. |
+| `lib.nlp.*` | 7 | **Nothing yet.** See the gap below. |
+| `lib.mail.*` | 3 | `gmail-search` / `gmail-add-label` / `gmail-archive`, over `@nodetool-ai/sandbox-nodetool/google`. |
+| `lib.secret.GetSecret` | 1 | `getSecret(name)` / `nodetool.secrets.get`. |
+| `lib.grid.CombineImageGrid` | 1 | `image.grid`, via the `image-grid` snippet. |
+
+Two of these reversed a decision recorded above, and both reversals turned on
+the same thing: the earlier audit asked whether a *guest script* could do the
+work, when the replacement was a **host capability**.
+
+- **`lib.mail`** was kept because "IMAP is not a `fetch` call". True, and beside
+  the point: `packages/agents/src/capabilities/email.ts` was already a second
+  implementation of all three nodes against the same `imapflow` client, so the
+  choice was never node-or-nothing, it was one implementation or two.
+- **`lib.secret.GetSecret`** was kept for its name picker. What decided it
+  instead was the gap between the node and the sandbox: the node read any secret
+  and fell back to `process.env[name]`, while `nodetool.secrets.get` is bound by
+  the run's declared `secretScope`. Keeping the node kept a way around a
+  guarantee the sandbox enforces.
+
+### The `lib.nlp` gap
+
+`lib.nlp` was removed with **no replacement**, deliberately and with the cost
+known. It took 3,267 lines with it — 501 of node class and 2,766 of vendored
+Porter/Lancaster/Spanish/French stemmers, a naive Bayes classifier, TF-IDF,
+phonetics and a sentiment lexicon — and no shipped example used any of it.
+
+Until a pack lands, stemming, TF-IDF, sentiment and phonetic matching have no
+in-product answer. Closing it means shipping `@nodetool-ai/sandbox-nlp` around
+`natural`: a config-only guest pack if it clears the QuickJS admission probe and
+the 1 MB cap, a host module otherwise. That is the follow-up; this round did not
+do it.

docs/cookbook/patterns.md

@@ -251,8 +251,8 @@ graph TD
 
 **Key Nodes**:
 
-- `GmailSearch` (`lib.mail.GmailSearch`): Search Gmail with queries
-- `Template`: Format email fields into text (lib.mail also provides `AddLabel` and `MoveToArchive`)
+- `Code` (`nodetool.code.Code`): Search Gmail and apply labels with `gmail_search` / `gmail_modify_labels` from `@nodetool-ai/sandbox-nodetool/google` — the `Gmail Search`, `Gmail Add Label` and `Gmail Archive` snippets prefill each one
+- `Template`: Format email fields into text
 - `Code` (`nodetool.code.Code`): Fetch web content with `fetch` and turn it into markdown with `@nodetool-ai/sandbox-html`, or parse an RSS feed with `@nodetool-ai/sandbox-xml` — there is no RSS node
 
 ______________________________________________________________________

docs/developer/node-patterns.md

@@ -257,23 +257,23 @@ The same approach applies to audio refs (see `audioBytesAsync` in `audio.ts`). T
 
 Use `@prop` with `type: "enum"` and a **`values`** array to present a dropdown in the UI.
 
-This pattern is taken from `TokenizeLibNode` in `packages/text-nodes/src/nodes/lib-nlp.ts`:
+This pattern is taken from `FilterStringNode` in `packages/text-nodes/src/nodes/text-extra.ts`:
 
 ```ts
 @prop({
   type: "enum",
-  default: "word",
-  title: "Mode",
-  description: "Tokenization mode: split into words or sentences",
-  values: ["word", "sentence"],
+  default: "contains",
+  title: "Filter Type",
+  description: "The type of filter to apply",
+  values: ["contains", "starts_with", "ends_with"],
 })
-declare mode: any;
+declare filter_type: any;
 ```
 
 In your `process()` method, read and cast the value from `this`:
 
 ```ts
-const mode = String(this.mode ?? "word");
+const filterType = String(this.filter_type ?? "contains");
 ```
 
 ---