@typeroll/mcp-server 0.45.41 → 0.45.43

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -230,11 +230,15 @@ link for your own verification:
230
230
  yourself) see a change.
231
231
  - **When they want the COMPILED static site** (a permanent bookmark, a
232
232
  stakeholder link to the built output, or a final pre-merge check): deploy
233
- the branch once (`trigger_deploy version="<branch>"`) and share the stable
234
- alias `https://<branch>.<project>.pages.dev` (the `<project>` is the part
235
- after the hash in the returned `deploy_url`). Branch deploys are
236
- `robots_blocked`, so it won't be indexed. The immutable per-deploy
237
- `<hash>.pages.dev` is for your own one-off checks (a new hash each deploy).
233
+ the branch once (`trigger_deploy version="<branch>"`) and share its stable
234
+ address: `deploy_url` from `list_versions` / `read_version` once the deploy
235
+ has succeeded. The publishing setup decides the host: under a Hosting Group
236
+ it is a `v-…` host under that group's site address base (which need not be
237
+ the site's own domain); otherwise it is the address the host reported. Never
238
+ construct it yourself, and do not guess a `<branch>.<project>.pages.dev`
239
+ alias. Branch deploys are `robots_blocked`, so the
240
+ address won't be indexed. A per-deploy `<hash>.pages.dev` URL in the job
241
+ details is for your own one-off checks (a new hash each deploy).
238
242
 
239
243
  Default: review/iterate on the reused DB-live preview; deploy only for the
240
244
  compiled output or merge. Wait for an explicit "looks good, ship it."
@@ -0,0 +1,71 @@
1
+ ---
2
+ name: tr-upgrade-rendering
3
+ description: Use when a site's platform render version is behind the latest (read_site_settings → render.version < render.latest), or the user asks to "upgrade rendering", "get the new block styles" or why a platform improvement doesn't show. Compares every page at the current and target version with screenshots, reports the differences and upgrades only with approval.
4
+ ---
5
+
6
+ # Upgrade a site's render version
7
+
8
+ Platform changes to block markup and shared CSS ship as numbered render
9
+ versions. A site keeps its version until someone upgrades it, so its look
10
+ never changes on its own. Upgrading is a single settings write, reversible by
11
+ writing the previous number. The comparison happens here, in the agent; the
12
+ server only renders previews.
13
+
14
+ The user can also do this in the portal: Settings → Rendering → "Preview with
15
+ version N", then "Upgrade". Offer this skill when they want a full-site visual
16
+ comparison first.
17
+
18
+ ## Recipe
19
+
20
+ 1. **Read the gap.** `read_site_settings` returns `render`:
21
+ `{ version, latest, upgrades: [{ version, title, changes[] }] }`. If
22
+ `version === latest`, stop. Tell the user what each pending version
23
+ changes, in their words, before measuring anything.
24
+
25
+ 2. **List the pages to compare.** `list_pages` (published and unlisted), plus
26
+ the header/footer, which appear on every page. For a large site, compare
27
+ the home page, one page per content type and template, and every page
28
+ whose blocks use a component named in `upgrades[].changes`.
29
+
30
+ 3. **Mint two links per page set.** One `get_preview_link` without
31
+ `render_version` (current) and one with `render_version: latest` (target).
32
+ Both show saved content. Mint each once and reuse it; internal links keep
33
+ the token.
34
+
35
+ 4. **Screenshot both** at 390, 768 and 1440 px wide, full page, after fonts
36
+ and images have loaded (`document.fonts.ready`, scroll to the end to
37
+ trigger lazy images, then back to the top). Use a headless browser with a
38
+ fixed viewport and `reducedMotion: 'reduce'`.
39
+
40
+ 5. **Compare.** Pixel-diff each pair (same size, small per-pixel tolerance)
41
+ and record the changed area and the bounding boxes of changed regions.
42
+ Crop each region from both versions. Ignore differences confined to
43
+ embedded third-party frames (maps, booking widgets, video).
44
+
45
+ 6. **Report** per page and width: unchanged, or the cropped before/after
46
+ regions with a one-line description ("links in body text are underlined",
47
+ "section gap 8px larger"). Link each change to the version note that
48
+ explains it. Say plainly when a change looks like a regression.
49
+
50
+ Before screenshots, read the site and page custom CSS for selectors a
51
+ version note names. Version 2 moves a heading's or button's custom class
52
+ onto the `<h2>` or link, so `.my-class .block-heading-text` stops
53
+ matching; propose the rewrite (`.my-class`) or a named style.
54
+
55
+ 7. **Ask before upgrading.** Never upgrade on your own. If the user wants to
56
+ keep a specific old look, solve it with a site style or page CSS first and
57
+ compare again.
58
+
59
+ 8. **Upgrade** with `update_site_settings render_version: <latest>` (pass
60
+ `version` to do it on a branch first). The preview changes at once; the
61
+ live site changes at the next deploy. To undo, write the previous number.
62
+
63
+ ## Pitfalls
64
+
65
+ - Comparing a draft against saved content shows content edits, not
66
+ rendering changes. Use the same content state for both links.
67
+ - Lazy images and web fonts cause false differences. Wait for them in both
68
+ screenshots.
69
+ - Comparing only the home page misses changes in components it doesn't use.
70
+ - Don't encode a workaround in custom CSS to resist an upgrade without
71
+ telling the user; it becomes a hidden fork of the platform styles.