jav-ed subtitle_Informer
subtitle_Informer is a preinstalled binary — just call it directly, no setup needed.
Read transcript content from subtitle files without flooding your context. Supports VTT (word-level and regular), SRT, and ASS. Always prefer the word-level file (en_Word.vtt) when available — it gives millisecond-precise timestamps per word instead of per phrase. Always start with info, then work from there:
subtitle_Informer info --file en_Word.vtt # total duration, word count, file type
subtitle_Informer chunks --file en_Word.vtt --words 500 # split into chunks — gives --from/--to boundaries to pass to slice
subtitle_Informer slice --file en_Word.vtt --from 1:00 --to 3:00 # returns actual timestamps + clean paragraph text
subtitle_Informer locate --file en_Word.vtt --text "they refused" # text → timestamps (copy text from slice output)
subtitle_Informer search --file en_Word.vtt --query "ceasefire" # find all mentions of a topic by keyword
Set once to skip --file on every call:
export TRANSCRIPT_SLICE_FILE=/path/to/en_Word.vtt
All commands support --json.
Commands
| Command | Input | Output |
|---|---|---|
info |
file | duration, word count, type |
chunks |
--words N |
--from/--to boundaries for each chunk |
slice |
--from, --to |
actual boundary timestamps + clean paragraph |
locate |
--text (copied from slice) |
timestamps — case-insensitive, what you see is what you search |
search |
--query (any keyword) |
all occurrences with timestamps and context |
locate vs search: use locate when you have text copied from slice. Use search when you are looking for a topic by keyword.
If You Need More
→ Full agent guide — file types, --overlap, --occurrence, --context-words, JSON schemas, typical workflows
See Also
→ content-creation skill — if you want to extract posts or shorts from what you're reading