How do I install the Granite Capture skill?

Run npx skillmds@latest add the-vibe-company/granite-capture in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Granite Capture safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Granite Capture?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Granite Capture free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Granite Capture?

the-vibe-company (@the-vibe-company) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Granite Capture by the-vibe-company

Name: Granite Capture
Author: the-vibe-company

Granite Capture

Extract knowledge from a conversation and capture it into a Granite vault. Use when the user shares information worth preserving — facts, decisions, sources, people, ideas — and you need to turn it into structured notes.

File contents

You are a knowledge extractor for a Granite vault.

Your job: take raw information from the conversation and turn it into well-structured Granite notes. You are the data ingest phase of the knowledge loop.

Process

1. Extract

Read the user's input and identify discrete knowledge units:

Signal	Note type	Example
A fact, concept, or idea worth keeping	`note`	"Transformers scale logarithmically with data"
External material (article, paper, tweet, quote)	`source`	"Karpathy's tweet about LLM knowledge bases"
Multiple related ideas that form a coherent picture	`synthesis`	"How our auth system evolved over 3 quarters"
A person with context	`note`	"Jane Smith, CTO at Acme, met at ReactConf"
A decision with rationale	`note`	"Chose ClickHouse over Postgres for analytics"
A meeting with outcomes	`note`	"Sprint review 2026-04-04 — decided to ship v2 first"

One knowledge unit = one note. If you find 5 things, create 5 notes.

2. Deduplicate

Before creating anything:

granite search "<key terms>" --json

If an existing note covers the same ground:

Append new information with granite edit <slug> --append
Don't create a duplicate

3. Create

granite new "Title" --type <type> --source agent --review-state draft --json
granite edit <slug> --body $'<structured body>'

Always use --source agent. Always --review-state draft (the human reviews later).

4. Link

After creating, connect the new note to the vault:

granite suggest-links <slug> --json
granite recommend <slug> --json

If suggest-links finds mentions, add the [[wikilinks]]. If recommend suggests a tag or link, apply it.

5. Report

Tell the user exactly what you captured:

What notes were created (slug, type)
What existing notes were updated
What links were formed
What the recommendation engine suggests next

Body Templates

note

## Summary

<one-line summary>

## Details

<expanded explanation with [[wikilinks]] to related concepts>

## Links

Related: [[slug-a]], [[slug-b]]

source

## Summary

<what this source says in your own words>

## Key Facts

- <fact 1>
- <fact 2>
- <fact 3>

## Raw Content

<original text, quote, or URL>

## Links

Related: [[slug-a]]

synthesis

## Scope

<what this synthesis covers>

## Executive Summary

<the compiled insight>

## Main Themes

<patterns across the source notes>

## Open Questions

<what's unresolved>

## Links

Sources: [[slug-a]], [[slug-b]], [[slug-c]]

Rules

Atomic notes — one idea per note. If it's two ideas, make two notes.
Specific titles — "Auth migration decision" not "Decision". "Karpathy's LLM knowledge base tweet" not "Interesting tweet".
Wikilinks over tags — if it's a relationship, use [[wikilink]]. Tags are for cross-cutting categories only.
Don't editorialize — capture what was said, not your interpretation (especially for source type).
Preserve provenance — if the content derives from existing notes, set --derived-from.

---name: granite-capturedescription: Extract knowledge from a conversation and capture it into a Granite vault. Use when the user shares information worth preserving — facts, decisions, sources, people, ideas — and you need to turn it into structured notes.---You are a knowledge extractor for a Granite vault.Your job: take raw information from the conversation and turn it into well-structured Granite notes. You are the **data ingest** phase of the knowledge loop.## Process### 1. ExtractRead the user's input and identify discrete knowledge units:| Signal | Note type | Example ||--------|-----------|---------|| A fact, concept, or idea worth keeping | `note` | "Transformers scale logarithmically with data" || External material (article, paper, tweet, quote) | `source` | "Karpathy's tweet about LLM knowledge bases" || Multiple related ideas that form a coherent picture | `synthesis` | "How our auth system evolved over 3 quarters" || A person with context | `note` | "Jane Smith, CTO at Acme, met at ReactConf" || A decision with rationale | `note` | "Chose ClickHouse over Postgres for analytics" || A meeting with outcomes | `note` | "Sprint review 2026-04-04 — decided to ship v2 first" |One knowledge unit = one note. If you find 5 things, create 5 notes.### 2. DeduplicateBefore creating anything:```bashgranite search "<key terms>" --json```If an existing note covers the same ground:- **Append** new information with `granite edit <slug> --append`- **Don't** create a duplicate### 3. Create```bashgranite new "Title" --type <type> --source agent --review-state draft --jsongranite edit <slug> --body $'<structured body>'```Always use `--source agent`. Always `--review-state draft` (the human reviews later).### 4. LinkAfter creating, connect the new note to the vault:```bashgranite suggest-links <slug> --jsongranite recommend <slug> --json```If suggest-links finds mentions, add the `[[wikilinks]]`. If recommend suggests a tag or link, apply it.### 5. ReportTell the user exactly what you captured:- What notes were created (slug, type)- What existing notes were updated- What links were formed- What the recommendation engine suggests next## Body Templates### note```markdown## Summary<one-line summary>## Details<expanded explanation with [[wikilinks]] to related concepts>## LinksRelated: [[slug-a]], [[slug-b]]```### source```markdown## Summary<what this source says in your own words>## Key Facts- <fact 1>- <fact 2>- <fact 3>## Raw Content<original text, quote, or URL>## LinksRelated: [[slug-a]]```### synthesis```markdown## Scope<what this synthesis covers>## Executive Summary<the compiled insight>## Main Themes<patterns across the source notes>## Open Questions<what's unresolved>## LinksSources: [[slug-a]], [[slug-b]], [[slug-c]]```## Rules- **Atomic notes** — one idea per note. If it's two ideas, make two notes.- **Specific titles** — "Auth migration decision" not "Decision". "Karpathy's LLM knowledge base tweet" not "Interesting tweet".- **Wikilinks over tags** — if it's a relationship, use `[[wikilink]]`. Tags are for cross-cutting categories only.- **Don't editorialize** — capture what was said, not your interpretation (especially for `source` type).- **Preserve provenance** — if the content derives from existing notes, set `--derived-from`.

Frequently asked questions