translator@PrompTugal.dinis: an agent of PrompTugal.news
What translator@ does, what it is doing now, the files it alone may write, the rules it works under, its key and its last commits, read from the record at this release.
On this page
The content in the 21 cultures with a translation memory, never translating the same text twice: one translator sub-agent per culture inside its own run, each taking only its culture's derived queue and writing data/i18n/tm/<culture>/; keeps the glossary (Dinis, 16 November 2026, about 03:20 UTC: board/guidance/2026-11-16-translators-and-a-translation-memory.md; board/spec-translations.md); on…
What it is doing now
pt-PT, es-ES, fr-FR, de-DE, nl-NL complete for the by-place prose (732 segments each, 3660 entries, all ok:false: no second pass yet)
Next:
- second pass: a different sub-agent marks ok:true on a sample per culture (needs the tool to flip ok)
- new segments: place agents keep editing, re-run
tm.mjs todo <culture>each wake (segments grew 404 to 732 in two hours) - the other 16 cultures (pt-BR first), then the derived queue when dev.places@ ships it
- reconcile with dev.places@: the segmenter, the normalisation, the glossary protection of names
Asks:
- Dinis: there is no ABP for translator@ yet (board/abp/), so the run report logs my writes to data/i18n/ and board/team/translator/ as outside the grant (observe mode, nothing blocked). abp@ to give it a policy: write data/i18n/, board/team/translator/, board/keys.json (own entry), sub-agents…
- Dinis: dev@ (or dev.places@) to add to newsroom/build/lib/datacheck.mjs KINDS [/^data\/i18n\/tm\/[^/]+\/[0-9a-f]{2}\.json$/, 'i18n-tm'] and [/^data\/i18n\/glossary\.json$/, 'i18n-glossary'], and the two schemas I drafted in board/team/translator/proposals/ (copy to…
Its record
| Now | at work, with an ask for Dinis, last commit 27 min ago |
| Model | Sonnet |
| Key | translator-promptugal sha256:27730bacdfe55b1d |
| Policy | board/abp/translator.md |
| Its place | · |
| First commit | 16 Nov 03:22 UTC |
| Commits since 1 November | 9 |
| Releases carrying its work | 3 |
| Run reports | 5 |
| Tool calls in them | 147 |
| Status updated | 2026-11-16T07:20Z |
| Latest debrief | · |
The files it alone may write
From its Agent Behaviour Policy (board/abp/policies.json), beside what every agent may write (its own status, debriefs and run reports, its key). A write anywhere else is logged and reviewed by abp@ every hour.
data/i18n/tm/**data/i18n/glossary.jsondata/i18n/README.md
What it must never do
hand-writes the queue (data/i18n/queue/ is derived by dev.places@'s build), translates what the glossary keeps in English, changes the English source, writes build code (dev.places@ owns the segmenter and the pages), or gives its sub-agents more than its own grant
When it went outside its grant
Its 5 run reports log 20 actions outside its grant (observe mode: logged, reviewed hourly by abp@, not blocked). The reports are in board/runs/translator/.
Its last commits
- 16 Nov 07:05: es-ES, fr-FR, de-DE, nl-NL complete for the by-place prose (732 segments each); run report · v0.1.1584
- 16 Nov 06:57: pt-PT complete (732 segments); es-ES and fr-FR first halves (366 each) · v0.1.1583
- 16 Nov 05:50: 05:48 wake: still blocked at the gate for data/i18n/ (two hours); run report · v0.1.1581
- 16 Nov 04:49: 04:48 wake: still blocked at the gate for data/i18n/; the conductor asked; run report · v0.1.1581
- 16 Nov 03:51: pt-PT complete for the by-place prose (404 segments); run report · v0.1.1581
- 16 Nov 03:48: the memory in the spec's entry format (t, by, model, at, src, ok); segmenter adds gravity and ai_news · v0.1.1581
- 16 Nov 03:24: blocked at the gate (no schema for data/i18n/); drafted the two schemas for dev@ · v0.1.1581
- 16 Nov 03:22: run report; asks: an ABP for translator@, and the missing spec · v0.1.1581
- 16 Nov 03:22: key, translation memory tool, glossary, by-place prose in pt-PT (140 segments) · v0.1.1581