altalandi 0.6.2 → 0.6.4
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/README.md +14 -0
- package/altalandi.mjs +32 -0
- package/batch.mjs +143 -20
- package/client.d.ts +238 -179
- package/client.mjs +288 -258
- package/package.json +1 -1
- package/report.mjs +20 -1
- package/skill/SKILL.md +17 -8
- package/update.mjs +1 -0
package/skill/SKILL.md
CHANGED
|
@@ -27,7 +27,7 @@ yourself or summarise the result in your own words.
|
|
|
27
27
|
line, below. It waits up to 5 minutes per call and handles long texts.
|
|
28
28
|
- **No shell, only the Altalandi tools** (`refine_icelandic`,
|
|
29
29
|
`proofread_icelandic`, `get_refinement_status`, for example in claude.ai or
|
|
30
|
-
ChatGPT): use the tools, and follow "Long texts
|
|
30
|
+
ChatGPT): use the tools, and follow "Long texts in parts" below. The
|
|
31
31
|
tools cannot wait as long as the command line, so long texts go in parts.
|
|
32
32
|
|
|
33
33
|
How long a call takes: a few seconds for a sentence, 10–30 seconds for a
|
|
@@ -73,21 +73,30 @@ A document of several sections (more than about 3,000 characters) goes better
|
|
|
73
73
|
through the folder run below: it refines and checks each passage on its own,
|
|
74
74
|
and a passage that fails keeps its original without holding up the rest. Save
|
|
75
75
|
the text as a `.md` file in an empty temporary folder and run the folder steps.
|
|
76
|
+
To send it in parts instead, follow "Long texts in parts".
|
|
76
77
|
|
|
77
|
-
## Long texts
|
|
78
|
+
## Long texts in parts
|
|
78
79
|
|
|
79
80
|
1. **Split the text into parts** at paragraph breaks, never inside a sentence:
|
|
80
81
|
at most **1,000 characters** a part for English, **1,500** for Icelandic.
|
|
81
82
|
Keep a heading with the paragraph under it. A list or a table stays in one
|
|
82
83
|
part when it fits.
|
|
83
|
-
2. **
|
|
84
|
-
|
|
84
|
+
2. **Command line: make one job id** for the whole text: `j_`, today's date
|
|
85
|
+
(yyyymmdd) and 6 random letters or digits, e.g. `j_20261013k4x9qa`. Run the
|
|
86
|
+
procedure once per part, in order, each with its own in and out files, and
|
|
87
|
+
add `--job <id> --part "Hluti N af M"` every time; on the first part add
|
|
88
|
+
`--job-title "<short name of the document>"` too. All parts are then one
|
|
89
|
+
verk on the site.
|
|
90
|
+
3. **Tools: call the tool once per part,** at most 3 at a time, in order. Give
|
|
91
|
+
every part its own new `operationId` (a fresh UUID). Keep a note of which ID
|
|
85
92
|
belongs to which part.
|
|
86
|
-
|
|
87
|
-
then each part's verdict and changes, as the tool returns
|
|
88
|
-
|
|
93
|
+
4. **Join the results in the original order** and give the user the whole text,
|
|
94
|
+
then each part's verdict and changes, as the command or the tool returns
|
|
95
|
+
them. With a job, end with its link once: `Verkið: <link>`, as the blocks
|
|
96
|
+
print it.
|
|
97
|
+
5. **A part that is refused or unverified keeps its original text.** Say which
|
|
89
98
|
part, and why (the tool says why). Never rewrite it yourself.
|
|
90
|
-
|
|
99
|
+
6. **Terms must match across parts.** If two parts translate the same term
|
|
91
100
|
differently, point it out and ask which to keep; do not quietly change one.
|
|
92
101
|
|
|
93
102
|
## If a call times out
|
package/update.mjs
CHANGED
|
@@ -142,6 +142,7 @@ export const SHIPPED_SKILLS = new Set([
|
|
|
142
142
|
"aeb88010599539fd", // 0.6.0
|
|
143
143
|
"ebfc4ef17a614973", // 0.6.1
|
|
144
144
|
"9def55ecc6b30785", // 0.6.2
|
|
145
|
+
"2168eaa25a5c561c", // after 0.6.2 (3c73702: stafir and krónur)
|
|
145
146
|
]);
|
|
146
147
|
|
|
147
148
|
export function skillFingerprint(text) {
|