docs: point restated upstream docs at live sections in the visual plugins

.changes/animation/upstream-pointers-visual-7a008454.md

@@ -0,0 +1,7 @@
+---
+bump: patch
+---
+
+### Changed
+
+- **The image long-edge limit is read live, not restated.** The rotoscope method's review record keeps our rule (tile and crop so no image is downscaled) and points at the vision guide's resolution table with an anchor, an as-of date and a recheck trigger; `learn-style` routes to that record instead of repeating the figure, and `review.py` keeps the value in force.

.changes/computer-use/upstream-pointers-visual-97c15a5b.md

@@ -0,0 +1,11 @@
+---
+bump: patch
+---
+
+### Fixed
+
+- **`/computer-use:diagnose` and `/computer-use:setup` no longer give an outdated Desktop settings path for enabling computer use.** The restated surface-differences table and eligibility list are replaced by our decision plus anchored pointers to the computer-use page's differences and troubleshooting sections and the Desktop page, with an as-of date and a recheck trigger.
+
+### Changed
+
+- **The screenshot guidance record points at the platform page's best-practices section, with the blog post kept as a correlate note.**

.changes/miro/upstream-pointers-visual-3688b257.md

@@ -0,0 +1,12 @@
+---
+bump: patch
+---
+
+### Fixed
+
+- **The vault-exec recipe's MCP pointers reach real sections.** The tool-name and deduplication basis named headings that do not exist; it now points at the MCP page's plugin-provided servers and scope-precedence sections.
+- **The vault-exec recipe's per-project `/mcp` disable step points at the section that records it.** Where `/mcp` stores a disabled-server choice is read from the MCP page's "Disable a server without removing it" section, not the scope-precedence section, which does not cover it.
+
+### Changed
+
+- **Restated Claude Code behavior in the README and `/miro:setup` becomes anchored pointer records.** Where sensitive option values are stored, when options are prompted for, the plugin CLI's scope defaults and flag spellings, and when Claude Code installs a plugin root's npm dependencies are read live. The setup bootstrap still passes the scope explicitly on every command; only its scope defaults and flag spellings are now read live instead of restated.

.changes/pixel-art/upstream-pointers-visual-49df9362.md

@@ -0,0 +1,7 @@
+---
+bump: patch
+---
+
+### Changed
+
+- **The brief's planning-offer record states our decision and points at the plugins overview.** The restated plugin-loading behavior is gone; the record keeps the rule (offer `/planning:interview` only when installed, say which path was taken) with anchored pointers, an as-of date and a recheck trigger.

.changes/playgrounds/upstream-pointers-visual-842bbe3d.md

@@ -0,0 +1,7 @@
+---
+bump: patch
+---
+
+### Changed
+
+- **`/playgrounds:use` reads skill-name resolution and plugin dependency behavior live.** The presence check keeps its decision (judge by install record, invoke the namespaced form) and the install step keeps its commands, while the mechanics behind them point at anchored sections of the skills, plugin dependencies and plugin install pages, with an as-of date and a recheck trigger.

.changes/playwright/upstream-pointers-visual-0bf0a71c.md

@@ -0,0 +1,11 @@
+---
+bump: patch
+---
+
+### Fixed
+
+- **The `demo-video` saved-value pointer reaches a live page again.** It named a retired docs slug; it now points at the plugins reference section that owns the topic.
+
+### Changed
+
+- **Restated upstream specifics become anchored pointer records.** The `check` skill's invocation-control record, the README's skill-size reasoning and the shared-login note's auto-mode observation now state our decision and point at the skills and permission-modes sections, with an as-of date and a recheck trigger.

.changes/prototype/upstream-pointers-visual-beab6111.md

@@ -0,0 +1,13 @@
+---
+bump: patch
+---
+
+### Fixed
+
+- **The `explore-directions` design record no longer lists an outdated artifacts version floor or an incomplete set of contexts where artifacts are off.** It points at the artifacts page's Availability and design-canvas sections instead, with an as-of date and a recheck trigger.
+
+- **The `detect-ecosystems.sh` wrappers in both skills no longer say `${CLAUDE_PLUGIN_ROOT}` stays literal in an `allowed-tools` rule.** They call the skill-local path a repo convention and point at the skills page's string-substitutions section for which variables a rule substitutes.
+
+### Changed
+
+- **The `design` and `ClaudeDesign` verification records name their probe and drop quoted upstream text.** Binary observations cite the records in `docs/native-surfaces/records.json`; `/design`'s behavior and the settings that remove it are read live from anchored sections.

.changes/visualization/upstream-pointers-visual-6b3bb5a7.md

@@ -0,0 +1,11 @@
+---
+bump: patch
+---
+
+### Fixed
+
+- **The visualize catalog no longer says a connector-backed page can never be shared publicly, and no longer lists an outdated artifacts version floor or an incomplete set of contexts where artifacts are off.** The connector, availability and design-canvas entries now state our decision and read the specifics live from anchored sections of the artifacts page, with an as-of date and a recheck trigger.
+
+### Changed
+
+- **Restated upstream specifics in `/visualization:visualize` and `/visualization:present` become anchored pointer records.** The design canvas's editing and export behavior, its settings and version gates, the `ClaudeDesign` write-refusal cases, the page-size and file-type limits, and the `userConfig` fixed-option floor are read live; the binary probe observations behind the `design` and `ClaudeDesign` records name `docs/native-surfaces/records.json` as their probe, and the mermaid runtime record names its probe pull request.

plugins/animation/skills/learn-style/SKILL.md

@@ -67,7 +67,7 @@ numbers live in `style.json` only; STYLE.md and other docs point at its keys.
    its distance to source: 0 is the source value, 1 the band edge on either side. The film's
    distance ranks passing scenes and its margin names the row nearest failing.
 6. Review: measure each prop with `--region X,Y,W,H --t T0-T1` next to source prop boxes, then
-   read every drawing at 1:1 (tile frames so no image is over the 2576 px long edge
+   read every drawing at 1:1 (tile frames so no image is over the long-edge limit
    `${CLAUDE_PLUGIN_ROOT}/skills/rotoscope/reference/method.md`, Reviewing images, records; never
    downscale, never a GIF) beside source crops. A pass on the numbers with a crop that does not
    read as the style is a fail; say which mark is wrong.

plugins/animation/skills/rotoscope/reference/method.md

@@ -100,12 +100,16 @@ quantized tone ramp.
 
 ## Reviewing images
 
-- Long-edge limit for any image read in review (crops, tiled frames). Claim: 2576 px on the long
-  edge is the largest image Claude sees without downscaling (Claude 4.7 and later; earlier models
-  1568 px), so a wider tile loses the 1:1 detail review needs. Basis: the Claude vision guide,
-  "Resolution and token cost" table (<https://platform.claude.com/docs/en/build-with-claude/vision>).
-  As of: 2026-09-26. Recheck trigger: that table changes, or review runs on a model outside it.
-  `review.py` sizes `CROP` from it (`LONG_EDGE`).
+- Long-edge limit for any image read in review (crops, tiled frames): tile and crop so no image
+  is larger than the reviewing model sees without downscaling, because a wider tile loses the 1:1
+  detail review needs. `review.py` holds the value in force as `LONG_EDGE` and sizes `CROP` from
+  it; when the pointer's table gives a different figure for the reviewing model, update that
+  constant.
+  - **Pointer**: when sizing a tile or crop, or when review runs on a new model, fetch
+    [Vision: Resolution and token cost](https://platform.claude.com/docs/en/build-with-claude/vision#evaluate-image-size)
+    live.
+  - **As of**: 2026-10-11
+  - **Recheck trigger**: that table changes, or review runs on a model outside it.
 
 ## Known limits
 

plugins/computer-use/README.md

@@ -45,9 +45,11 @@ applies there, but **no macOS quirks file ships because none has been verified.*
 generate plausible macOS specifics, the plugin states the gap and serves macOS users the
 platform-neutral material only.
 
-Surfaces are not interchangeable, and the skill says so: computer use in the Claude Code CLI is
-macOS-only, while the Desktop app covers macOS and Windows. Guidance written for one is wrong on
-the other.
+Surfaces are not interchangeable, and the skill says so: the Claude Code CLI and the Desktop app
+differ in the platforms they run computer use on and in how they enable it, so guidance written
+for one is wrong on the other. The skills read that split live from
+[Computer use: Differences from the Desktop app](https://code.claude.com/docs/en/computer-use#differences-from-the-desktop-app)
+(as of 2026-10-11; recheck when that section's table changes).
 
 ## What it deliberately does not do
 
@@ -61,6 +63,9 @@ the other.
 
 ## Prerequisites
 
-The `computer-use` MCP server must be enabled, on a plan that supports it (research preview; Pro
-or Max, not Team or Enterprise), with OS-level permissions granted. `/computer-use:setup` probes
-all of it and reports what is missing; installing and enabling remain the operator's.
+The `computer-use` MCP server must be enabled, on an eligible setup, with OS-level permissions
+granted. `/computer-use:setup` probes all of it and reports what is missing; installing and
+enabling remain the operator's. For which plans, providers and session modes are eligible, see
+[Computer use: Troubleshooting](https://code.claude.com/docs/en/computer-use#troubleshooting)
+(as of 2026-10-11; recheck when that section changes the plans, providers or session modes it
+names).

plugins/computer-use/skills/diagnose/SKILL.md

@@ -58,20 +58,17 @@ verified. A macOS user gets the platform-neutral material and no fabricated spec
 
 ## Surfaces differ. Know which one you are on
 
-Computer use is not one thing, and guidance written for one surface is wrong on the other. The rows
-below restate the "Differences from the Desktop app" table on [computer use from the
-CLI](https://code.claude.com/docs/en/computer-use), verified 2026-09-06 against Claude Code 2.1.263
-and that page as fetched that day. Recheck when that page drops the differences table, when a row in
-it changes, or when a release note names computer use on either surface.
-
-| | Claude Code CLI | Claude Desktop |
-|---|---|---|
-| Platforms | macOS only | macOS **and** Windows |
-| Enabling | `/mcp` → enable `computer-use` | Settings → General |
-| Denied-apps list | not available | configurable |
-
-So a Windows session using computer use is on the **Desktop** surface by construction. Check
-before applying any surface-specific advice.
+Computer use is not one thing, and guidance written for one surface is wrong on the other. Name
+the surface (Claude Code CLI or Claude Desktop) before applying any surface-specific advice. Which
+platforms each surface runs on, how each one enables computer use, and which settings each one
+offers are the upstream page's to state; read them there rather than from this skill.
+
+- **Pointer**: when the surface, its platforms or its enabling path matters to a diagnosis, fetch
+  [Computer use: Differences from the Desktop app](https://code.claude.com/docs/en/computer-use#differences-from-the-desktop-app)
+  live.
+- **As of**: 2026-10-11
+- **Recheck trigger**: that section's table adds, drops or changes a row, or a release note names
+  computer use on either surface.
 
 ## Gotchas
 

plugins/computer-use/skills/diagnose/reference/screenshots-and-zoom.md

@@ -93,6 +93,13 @@ translate the API advice into a display-settings change on a Claude Code machine
 - **Recheck trigger**: either section changes its recommended resolutions, or the Claude Code
   computer-use page documents a setting to change the target size.
 
+The wider implementation advice on the same page has its own record:
+
+- **Pointer**: for the wider implementation advice, see
+  [Follow implementation best practices](https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool#follow-implementation-best-practices).
+- **As of**: 2026-10-11
+- **Recheck trigger**: that section changes its screenshot, resolution or click-accuracy advice.
+
 ## `save_to_disk` is not an escape hatch
 
 `screenshot` accepts `save_to_disk: true`, but on Windows no file was found anywhere under the

plugins/computer-use/skills/setup/SKILL.md

@@ -27,11 +27,15 @@ run them and report. Emit a PASS / FAIL / INFO table with one remediation line p
 
 Determine which surface is in play, because the answer changes what is even possible:
 
-- **Windows** → necessarily the Claude Desktop surface; the CLI's computer use is macOS-only
-  (CLI computer-use page, <https://code.claude.com/docs/en/computer-use>, verified 2026-09-02;
-  recheck when that page's "Differences from the Desktop app" table changes).
-- **macOS** → either surface; check whether `computer-use` is enabled in `/mcp` (CLI) or in
-  Settings → General (Desktop).
+- Read which platforms each surface runs on and where each one enables computer use from the
+  pointer below, then match the session's operating system and enabling path against it. When the
+  operating system leaves one surface possible, that is the surface; otherwise check the enabling
+  path the pointer names for each surface.
+  - **Pointer**: when deciding the surface, fetch
+    [Computer use: Differences from the Desktop app](https://code.claude.com/docs/en/computer-use#differences-from-the-desktop-app)
+    live.
+  - **As of**: 2026-10-11
+  - **Recheck trigger**: that section's Platforms or Enable row changes.
 
 Report the surface as INFO. It is context for everything below, not a pass/fail.
 
@@ -43,10 +47,15 @@ allowlist and the active grant flags.
 
 - Tools reachable → PASS, and report the current allowlist and grant flags as INFO.
 - Tools absent → FAIL. Remediation: enable the `computer-use` MCP server for this project, and
-  confirm the session is eligible: research preview on a Pro or Max plan (not Team or Enterprise),
-  authenticated through claude.ai rather than a third-party provider, and an interactive session
-  (not `-p`). Eligibility per the CLI and Desktop pages, verified 2026-09-02; recheck when either
-  page's eligibility note changes.
+  confirm the session meets each eligibility condition the pointer lists, naming each unmet one
+  in the report.
+  - **Pointer**: when the tools are absent, fetch
+    [Computer use: Troubleshooting](https://code.claude.com/docs/en/computer-use#troubleshooting)
+    live for the CLI's eligibility checks, and
+    [Desktop: Let Claude use your computer](https://code.claude.com/docs/en/desktop#let-claude-use-your-computer)
+    for the Desktop surface.
+  - **As of**: 2026-10-11
+  - **Recheck trigger**: either section changes which plans, providers or session modes it names.
 
 Do **not** call `request_access` during a check. That raises a consent dialog, which is not
 read-only behavior.