@kolbo/mcp 1.88.1 → 1.88.2
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.
package/package.json
CHANGED
package/skill/GENERATED.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# AUTO-GENERATED — do not edit
|
|
2
2
|
|
|
3
|
-
This tree is mirrored from kolbo-code@
|
|
3
|
+
This tree is mirrored from kolbo-code@0ac4fba, the single source of truth.
|
|
4
4
|
Canonical source: packages/opencode/skills/kolbo/
|
|
5
5
|
Distribution: .github/workflows/sync-skill-to-plugin.yml
|
|
6
6
|
|
package/skill/SKILL.md
CHANGED
package/skill/VERSION
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
0.9.
|
|
1
|
+
0.9.15
|
|
@@ -34,6 +34,31 @@ The backend renders internal specimens. Do NOT render specimens, attach them as
|
|
|
34
34
|
|
|
35
35
|
Font upload/preparation has no separate credit charge; normal image generation remains billable under the existing approval rules. No automatic paid regeneration to improve typography.
|
|
36
36
|
|
|
37
|
+
## Making the typeface actually come through
|
|
38
|
+
|
|
39
|
+
Measured 2026-09-10 by rerunning one customer ad (Hebrew, custom family, three reference
|
|
40
|
+
images) across models and settings. The backend renders a specimen and the model IMITATES
|
|
41
|
+
it — nothing installs the font — so these are the levers that decide how close it lands.
|
|
42
|
+
|
|
43
|
+
- **Model choice is the biggest one.** GPT Image 2 reproduced the uploaded letterforms
|
|
44
|
+
clearly better than GPT Image 2.5 Sunburst / Flare, which drift toward a default bold
|
|
45
|
+
Hebrew. Recommend GPT Image 2 whenever the typeface matters.
|
|
46
|
+
- **Quality does not compensate.** 2K + `high` on GPT Image 2 beat both 2.5 rows at
|
|
47
|
+
`max`. Do not sell a higher tier as a fix for typography.
|
|
48
|
+
- **Weight words in the prompt beat the specimen.** "bold", "medium weight", "very large
|
|
49
|
+
bold headline" read as typeface instructions and usually win — an ad that said bold five
|
|
50
|
+
times came back in a generic sans. Coach the user to describe size, placement, colour and
|
|
51
|
+
glow, and to choose the weight by selecting the uploaded STYLE (Bold / Medium / Light)
|
|
52
|
+
instead of writing it. Keep their exact-copy line ("EXACTLY letter for letter").
|
|
53
|
+
- **Busy layouts drift; calm ones do not.** The same font on a simple prompt reproduced
|
|
54
|
+
almost exactly, and on a split-screen ad with three competing references it was ignored.
|
|
55
|
+
Fewer competing reference images and fewer text blocks buy real fidelity.
|
|
56
|
+
- **Emoji never block a generation** and are drawn from the platform emoji set; they are
|
|
57
|
+
excluded from the specimen by design. No font carries them.
|
|
58
|
+
- **When it must be exact, say so.** For client-final work where the typeface cannot drift,
|
|
59
|
+
generate the layout with the text areas empty and set the type over it. Never promise
|
|
60
|
+
faithful reproduction — the model is imitating a picture of the letters.
|
|
61
|
+
|
|
37
62
|
## SDK / REST
|
|
38
63
|
|
|
39
64
|
The account-authenticated server-side SDK exports `createFontClient`: `list`, `get`, `upload(Blob, filename)`, `status`, `rename`, `delete`, `createUploadTicket`, and `grantToApp`. Keep account API keys on the server. The dedicated REST root is `/api/v1/fonts`; multipart upload is POST to that root. App end-user credentials do not grant access to an owner's personal library; use explicit app font grants. Image SDK calls use the same optional `font_ids`.
|