luxin-cli 0.2.3 → 0.2.5
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/CHANGELOG.md +61 -0
- package/bin/luxin.mjs +830 -206
- package/cli.md +142 -52
- package/commands.json +1 -0
- package/llms.txt +6 -6
- package/package.json +2 -3
- package/skill.md +14 -10
- package/skills/agent-image-generation/SKILL.md +1 -1
- package/skills/ai-audio-generation/SKILL.md +1 -1
- package/skills/ai-image-generation/SKILL.md +1 -1
- package/skills/ai-video-generation/SKILL.md +1 -1
- package/skills/creative-media/SKILL.md +1 -1
- package/skills/image-edit/SKILL.md +1 -1
- package/skills/image-generation/SKILL.md +2 -2
- package/skills/image-skill/SKILL.md +78 -0
- package/skills/image-to-3d/SKILL.md +2 -2
- package/skills/luxin/SKILL.md +14 -10
- package/skills/luxin/references/cli.md +142 -52
- package/skills/luxin/references/commands.json +1 -0
- package/skills/luxin/references/llms.txt +6 -6
- package/SKILL.md +0 -436
package/CHANGELOG.md
CHANGED
|
@@ -6,6 +6,67 @@ provenance; this file is the human- and agent-readable release map.
|
|
|
6
6
|
|
|
7
7
|
## Unreleased
|
|
8
8
|
|
|
9
|
+
## 0.2.5 - 2026-08-09
|
|
10
|
+
|
|
11
|
+
- Release (correctness): `edit --guide` selects a model again. It gated on
|
|
12
|
+
`media.input.images.required`, which the hosted `/v1/models` summary does not
|
|
13
|
+
publish, so on 0.2.4 every edit brief returned `no_executable_model` and the
|
|
14
|
+
blocker named `create`. The predicate now reads `supports`, which the summary
|
|
15
|
+
does publish; measured on the live 220-model catalog the two express the same
|
|
16
|
+
79-model set, so this drops a redundant conjunct rather than a guarantee. The
|
|
17
|
+
blocker also names the operation it actually failed on.
|
|
18
|
+
- Release (correctness): a plain video brief reports its aspect ratio again.
|
|
19
|
+
`selection.suggested_aspect_ratio` read the same unpublished block and
|
|
20
|
+
returned null on every real video call unless the brief happened to name a
|
|
21
|
+
ratio.
|
|
22
|
+
- Known issue (#2259): `create --guide` can quote a different credit figure than
|
|
23
|
+
the `--dry-run` it hands you, because it prices the model's catalog default
|
|
24
|
+
shape rather than the ratio the command actually sends. Measured against the
|
|
25
|
+
live API: 4 vs 5 with no ratio stated, 4 vs 3 at 16:9. The dry-run figure is
|
|
26
|
+
the authoritative one; budget from it, not from the guide. Unchanged from
|
|
27
|
+
0.2.4; the fix landed server-side but not in this package.
|
|
28
|
+
|
|
29
|
+
## 0.2.4 - 2026-08-09
|
|
30
|
+
|
|
31
|
+
- Release (correctness): the guide honors an aspect ratio stated in the brief
|
|
32
|
+
instead of dropping it (#2213, #2203, #2209). A "16:9 cinematic widescreen
|
|
33
|
+
poster" brief previously returned a 2048x2048 square and charged 12 credits
|
|
34
|
+
for it: `selection.suggested_aspect_ratio` was hard-null for every image
|
|
35
|
+
model because the guide only ever suggested a ratio for video. The guide now
|
|
36
|
+
extracts an explicit ratio conservatively, treats it as a required capability
|
|
37
|
+
during model selection the way #2225 treats seed, and passes
|
|
38
|
+
`--aspect-ratio` through to create, dry-run and edit. Bare "portrait" and
|
|
39
|
+
"landscape" deliberately do not fire, and neither do quantity ratios ("the
|
|
40
|
+
ratio of sugar to flour is 3:2") or bare clock times.
|
|
41
|
+
- Release (output): `create --guide` at an unblocked `ready_to_create` returns
|
|
42
|
+
the decision rather than every field (#2204 and its cluster). Measured
|
|
43
|
+
21,667 to 7,911 bytes, 449 to 200 lines. A blocked agent still receives the
|
|
44
|
+
full recovery and self-fund apparatus, because that is when it is the right
|
|
45
|
+
answer, and the pre-wall funding nudge survives whenever it is genuinely
|
|
46
|
+
recommended. `data.output_mode` names every omitted and compacted field and
|
|
47
|
+
carries a copy-runnable `--explain`, so nothing is trimmed silently.
|
|
48
|
+
- Release (cost): `--intent` now drives model selection (#2231). `draft`
|
|
49
|
+
previously selected a 25-credit model against a 12-credit default and now
|
|
50
|
+
selects a 1-credit one; the unhinted default drops from 12 to 4 credits.
|
|
51
|
+
`final` still selects the quality model, and reproducibility briefs still
|
|
52
|
+
route to a seed-capable model. `cost.cheaper_alternative` names one available
|
|
53
|
+
model that costs strictly less and still satisfies the brief.
|
|
54
|
+
- Release (trust): public responses no longer publish our provider cost,
|
|
55
|
+
revenue, realized margin, target margin, or payment-fee model. What a caller
|
|
56
|
+
is charged is unchanged; what we paid and made is no longer disclosed.
|
|
57
|
+
- Release (dry-run): hosted create and edit responses carry an explicit
|
|
58
|
+
`data.mutation` stating `provider_call`, `credit_debit` and `media_write`
|
|
59
|
+
rather than leaving them to be inferred (#2216), and planned assets are
|
|
60
|
+
labelled as placeholders rather than media.
|
|
61
|
+
- Release (contract): `aspect_ratio` is documented in `skill.md` as the
|
|
62
|
+
top-level normalized control it always was.
|
|
63
|
+
- Release (media): newly created assets are returned at `media.luxin.sh`.
|
|
64
|
+
Existing `media.image-skill.com` URLs continue to resolve and are still
|
|
65
|
+
accepted as input.
|
|
66
|
+
- Release (discovery): the `image-skill` skill slug resolves again. The rename
|
|
67
|
+
removed it and `npx skills add ... --skill image-skill` had failed since
|
|
68
|
+
2026-07-13 on the mirror path carrying most of our installs.
|
|
69
|
+
|
|
9
70
|
## 0.2.3 - 2026-08-04
|
|
10
71
|
|
|
11
72
|
- Release (self-fund/recovery): the wall now offers a human-funding handoff
|