| name | save |
| description | Research and save a book, video, blog post, article, paper, podcast, link, or other resource into memory. Use when the user says "/save", "save this", "remember this", "store this resource", or asks Claude to quickly research something and log it into `~/Dropbox/memory/resources/`. |
save
Quickly research a resource and store a durable note in memory.
Workflow
-
Parse the user's input into a specific resource.
- Accept a title, a URL, or both.
- If the user gives only a vague topic instead of a specific item, ask one short clarifying question.
-
Do lightweight research.
- If the user included a URL, fetch/read it first.
- If the user named a book, video, blog post, article, paper, or similar item without a URL, use the best available tools to identify the canonical source and gather a quick summary.
- Keep the research pass short. Aim for a concise, high-signal summary rather than exhaustive notes.
- If browsing tools are unavailable, use the best available local context or model knowledge and say so briefly.
-
Normalize the note before saving.
- Infer
type as one of: book, video, blogpost, article, paper, podcast, course, talk, link, other.
- Capture:
title
type
source if known (prefer the canonical URL; url is acceptable as a legacy alias)
tags (optional Obsidian tags as a list; omit # prefixes)
summary (2-6 bullets or a short paragraph)
why_saved (why this matters to the user right now)
research_notes (optional extra bullets)
author (optional note author; defaults to Claude)
-
Write the note using the bundled helper script scripts/write_resource_note.py.
- The script writes one markdown file per resource under
~/Dropbox/memory/resources/<bucket>/.
- Prefer this script over hand-writing the file so filenames and sections stay consistent.
-
Refresh memory indexing.
- If
qmd is available, run qmd update after saving.
-
Report back succinctly.
- Tell the user what you saved.
- Include the final file path.
- Include a one-line summary of the resource.
Helper Script
Use scripts/write_resource_note.py with JSON on stdin.
Example:
python3 scripts/write_resource_note.py <<'EOF'
{
"title": "Deep Work",
"type": "book",
"source": "https://example.com",
"tags": ["books", "productivity", "focus"],
"summary": "A book about focused, distraction-free work as a competitive advantage.",
"why_saved": "Useful for thinking about focus and execution.",
"research_notes": [
"Emphasizes reducing shallow work.",
"Frames concentration as a trainable skill."
]
}
EOF
type is the preferred field name. The helper still accepts kind as a legacy alias.
Output Format
Save notes in this shape, with YAML frontmatter first:
---
title: Deep Work
created: 2026-04-19
type: book
source: https://example.com
tags:
- books
- productivity
- focus
author: Claude
---
# Deep Work
## Summary
Short summary.
## Why saved
Why it matters.
## Research notes
- Bullet
- Bullet
Frontmatter Template
Use this template when saving a resource note:
---
title: <resource title>
created: <YYYY-MM-DD>
type: <book|video|blogpost|article|paper|podcast|course|talk|link|other>
source: <canonical url or source path, if known>
tags:
- <tag1>
- <tag2>
author: Claude
---
Rules
- Save into
~/Dropbox/memory/resources/, not links/, when the resource is something the user wants to remember with a summary.
- Keep summaries crisp and durable.
- Do not invent details when the source is unclear; say what is uncertain.
- Prefer one file per resource.
- If the same resource already exists, update or append to the existing note instead of creating a duplicate.