Hanabank's users find this name easier to understand. Display text changes in
English and Indonesian, and "1.2" goes with it. The package format, internal
ids and schedule names stay as they are.
The chat's example now says INTERACTIVE, so the parser reads that word as well.
A bare "interactive" only counts when no other kind is named, so "kuis
interaktif" stays a quiz.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Generic requests ("10 konten", "10 pdf") now become a rule whose kind the
agent asks about, and a rule with no frequency is asked about instead of
silently becoming weekly. Times like 09.00 no longer split sentences, and
requests about drafts, the hour, scheduling, sources and duplicates get a
truthful note rather than an "I did not understand" loop.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Brain and AI page lists Brain material (failed documents hidden) with
search, open, delete and per-material or select-all ticks. New schedules use
only the ticked material, kept through topic edits and resumes; nothing ticked
means the whole Brain. With material ticked, the agent skips the content topic
question and names the material in the chat and plan.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The agent answers in the language each message is written in (Indonesian
or English), whatever the UI language; short replies such as "LLM" keep
the conversation's language. inLang() runs t() in that language.
Each turn now reads like a person: a thinking step with what the agent is
doing, the reply typed out with "Thought for Ns" above it, a short typing
pause, then one question, with its options shown once it is typed.
Reduced motion skips the animation.
"dari file LLM" / "from the document X" is read as the topic, and topics
keep the user's capitals.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Instruction agent now confirms the output language before the plan is
ready: Bahasa Indonesia, English or 中文, the languages the engine writes
in. It can also be named in the instruction ("pakai bahasa Inggris"), is
read from short replies, and can be changed in the plan. Research alone
is not asked about.
The engine's content production has no language field, so the choice
goes into the agent's instructions; video schedules also set
video_options.language. ruleOf reads it back, so editing a topic keeps
it. The audio player recognises Chinese scripts and uses a zh-CN voice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Instruction page is now a conversation with the Express Agent. Each
message updates the draft plan, and when something is unclear (a part it
did not understand, a video count with no cadence, a schedule over 50 per
week, weekly research with no day, a missing topic) the agent asks right
in the chat, one question at a time, with replies to pick or type. Once
the plan is clear it offers to start the schedules and reports the result
in the chat. The plan stays editable beside the chat.
The chat text is also larger and darker than the mockup's 9-11px grey.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Includes fixes found while reviewing the app:
- Results: open each result by a type+agent+id key, so a video and a
material draft with the same id no longer open each other's popup
- AuthGate: clear the query cache during render, before pages build their
queries, and remount pages on organization switch (pages no longer stay
on skeletons for up to a minute after sign-in or refresh)
- Instruction: keep only failed rules after a partial approve, so retrying
does not create the successful schedules twice; drop the monthly note
once count or frequency is edited
- Parser/schedules: never schedule a count below one
- Audio: ignore the end event of a cancelled utterance when changing speed
- Downloads: revoke the object URL later so Safari/Firefox can read it
- Results: refetch SCORM content when any part's revision changes; show a
failed video as failed instead of "still rendering"
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
