ingest-document
/ingest-document
Dołącza źródło do librocat, np. PDF, arkusz, slajdy lub nagranie, i tworzy z niego pojęcia z cytowaniami.
- Autor
- laughing-hermit ↗
- Repozytorium
- librocat ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 10.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Dołącza źródło do librocat, np. PDF, arkusz, slajdy lub nagranie, i tworzy z niego pojęcia z cytowaniami.
Wyzwalacze z opisu
Oryginalny opis (EN)
Attach a source to librocat (a PDF, a document, a spreadsheet, slides, a picture, a web page, a video or podcast's transcript) or read a code repository, then turn it into concepts that cite it. Use when the user points at a file, a URL, or a repo ("attach this", "read this doc / page / video and keep what matters", "catalog this repo"), and for files people attached on the dashboard that no concept cites yet.
Jak działa
- Attach the source
- Reading each kind of source
- Procedure
Podgląd SKILL.md
---
name: ingest-document
description: >-
Attach a source to librocat (a PDF, a document, a spreadsheet, slides, a
picture, a web page, a video or podcast's transcript) or read a code
repository, then turn it into concepts that cite it. Use when the user
points at a file, a URL, or a repo ("attach this", "read this doc / page /
video and keep what matters", "catalog this repo"), and for files people
attached on the dashboard that no concept cites yet.
---
# Attach, then catalog
First attach the source: librocat keeps the original and reads its text
where code can, so any agent finds it by its words and its title at once,
even before anyone writes about it. Then catalog it: read it and write what
matters as concepts that cite it. The rules for those concepts (what
belongs, one concept per idea, the types, the catalog record, acting on
`possible_duplicates`) are the **librocat-write** skill's; follow it.
librocat keeps text and pictures only. For a recording, keep its
transcript as text, never the audio or video file.
## Attach the source
- **A file, with a shell:** `add_source` with `file` (name, type, size in
bytes). POST the bytes to the `upload_url` it returns
(`curl -X POST -H 'Content-Type: <type>' --data-binary @<file> '<upload_url>'`
answers `{"storageId": …}`), then `add_source` with that `storage_id`,
the file's name, and a title. librocat reads PDFs and Office files itself.
- **A file, without a shell** (some desktop agents can't send the bytes):
ask the person to attach it on the dashboard (the Library's Attach files), then find it
with `get_source` (no arguments: files no concept cites come first). Or
keep the text you read with `add_source`, `title` and `text`.
- **A web page:** `add_source` with its text, its title, and its URL as
`location`. A PDF, an Office file, or a picture at a URL: on paid plans, `add_source`
with only the URL as `location` keeps the file itself; otherwise
download it and upload it as a file. A site that doesn't let apps read
it (LinkedIn, X): read the page yourself if you can, and keep its text
with the URL as `location`.
- **Text you already have** (a pasted note, a transcript): `add_source`
with the title, the text, and the `location`.
- **A picture or a scan** has no words to find: give it a title that says
what it shows.
- **"Save this for later"** (paid plans): `add_source` with only the URL as
`location` keeps the page on the reading list, unread until a concept
cites it.
- A file reads `status: processing`, then `ready` within a minute;
`get_source` shows what librocat read. Adding the same file or text again
returns the kept copy; new text for a kept URL replaces it, and the
concepts citing it come up for review.
- Skip this for code in a repository (the repository is the source) and
for anything the user doesn't want kept.
## Reading each kind of source
- **Documents** (PDF, Word, a spreadsheet, slides, Markdown, meeting notes,
a spec): note the title, author, date, and path or URL.
- **Web pages**: read the article, not the navigation, ads, or comments.
Keep the canonical URL (no tracking parameters) and the page's date.
- **Videos and podcasts**: work from a transcript (the page's captions, the
show notes, or one the user gives you), never from a title alone. If none
is within reach, say so and ask for one. Cite timestamps (`12:34`), and
link to the moment where the platform allows it (YouTube's `&t=754s`).
- **Code repositories**: one `Note` per module or
component, with `sources` naming its files, and the decisions and
runbooks the files don't state. A repository's own docs (README, `docs/`,
ADRs) are documents. After a commit, `files_changed` brings the concepts
that cite the changed files up for review.
Read only what the user can already read: never work around a paywall or a
sign-in.
## Procedure
1. Attach the source (above) and note its title, author, date, location,
and the returned `cite`.
2. List the candidate concepts (idea, type, one-line description). Show the
list to the user if there are more than ten, and let them cut.
3. `search` each one's key terms: new concept, `update`, or skip. Then
look for what the source contradicts: `search` each fact it states that
can go stale (a price, a version, an owner, a date, a limit, a step),
and `update` the concepts it changes, citing it, or `propose` the change
when the person should decide.
4. `ingest` them in one call with `concepts: [...]`. Each gets the full
catalog record, the `cite` in `sources`, and a body in your own words
that ends with a `## Source` section: title, author or channel, date,
URL or path, and the timestamps it draws on. Quote only the short lines
that must be exact (numbers, names, commands, contract terms).
5. Act on any `possible_duplicates` (librocat-write).
6. Report how many concepts you wrote, their ids, the `cite`, and what you
left out.
When a question later needs the exact wording, `get_source` with the
concept's `sources` entry and a `query` returns the passages.
Wersje
W bazie nie ma innych skilli o nazwie ingest-document.
Migawka bazy z 13.11.2026 · wersja paczki 9817bc184248 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90librocat-write
Pokazuje, co wpisywać do biblioteki librocat: jedno pojęcie na ideę, typy pojęć i pełny rekord z tagami, źródłami i datą przeglądu. - 100ingest
Zamienia surowe źródło (wycinek, artykuł, transkrypcję) w uporządkowaną stronę pojęcia w wiki vaulta i rejestruje ją w indeksie oraz logu. - 100wiki-ingest
Dodaje nowe źródło do wiki, na przykład artykuł, pracę, transkrypt lub plik, i aktualizuje powiązane strony wiki. - 100ingest
Wczytuje jeden materiał, czyli PDF, stronę lub plik, do wiki LLM w notatniku, zapisuje strony źródłowe, indeks i dziennik. - 100read
Czyta jedno zewnętrzne źródło, czyli stronę, wpis, wideo, repozytorium, PDF lub plik, i przenosi jego treść do czatu bez zapisywania. - 90attach
Dołącza plik do właściwej notatki AIOS według poziomu ekspozycji: lokalnie bez wysyłania albo w zabezpieczonym magazynie.