trove-ingest

Original：🇺🇸 English

Translated

Use when the user shares a source (URL, file path, paste, transcript) that should be indexed into the Trove memory graph. Stores the raw evidence first via ingest, then remembers semantic atoms citing exact text units.

2installs

Sourceanunay999/trove

Added on2026-08-26

NPX Install

npx skill4agent add anunay999/trove trove-ingest

SKILL.md Content

View Translation Comparison →

trove-ingest

Evidence first, meaning second. The graph never contains a claim it cannot prove.

When to use

User shares a URL, file, gist, or paste worth remembering.
A long document (spec, transcript, article) should become queryable memory.

When not to use:

Conversation-born knowledge with no external source →
```
trove-remember
```
.
The source is a vault page — the importer handles those (
```
npm run import:vault
```
).

Process

Step 1 — obtain the text

Fetch/read the source. For URLs, extract the readable text.

Step 2 — store the evidence

ingest { kind, title, uri, contentText, metadata }

```
kind
```
:
```
url
```
|
```
file
```
|
```
paste
```
|
```
transcript
```
|
```
email
```
|
```
slack
```
.
Identical content dedupes by hash — re-ingesting is safe.
Returns the
```
source
```
and its addressable
```
textUnits
```
.

Step 3 — extract atoms

From the text units, remember the meaning-bearing atoms (typically far fewer than the text is long): entities, claims, decisions, patterns. For each:

remember { title, type, summary, evidence: [{ textUnitId }], links }

Every atom cites the exact text unit that supports it. Link atoms to existing project/domain nodes via

links

; remember's response lists similar nodes if the target already exists.

Step 4 — annotate (optional)

annotate

to mark important spans ("contradicts claim X", "important quote") without rewriting anything.

Step 5 — confirm

Source id + how many text units; atoms remembered with slugs; anything that contradicted existing nodes (recommend

trove-remember

or

trove-lint

).

Anti-patterns

Don't remember atoms without text-unit citations when the source is right there.
Don't turn the whole document into one giant node — the source already stores the full text; atoms are the compression.
Don't ingest secrets or credentials as content.

trove-ingest

NPX Install

Tags

SKILL.md Content

trove-ingest

When to use

Process

Step 1 — obtain the text

Step 2 — store the evidence

Step 3 — extract atoms

Step 4 — annotate (optional)

Step 5 — confirm

Anti-patterns