Skip to content
Back to skills

Whiteboard Curate Suggest

ASecurity

When an insight, decision, or checklist with lasting value comes up during a conversation, suggest saving it to the library, and on the user's consent register it in Curated Knowledge via knowledge_promote.

  • 11 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 19, 2026
ai-agentsrustgoexpress

Security analysis

A100/100

Scanned September 19, 2026

npx -y skills add CocoRoF/Geny --skill whiteboard_curate_suggest --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Whiteboard Curate Suggest?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Whiteboard Curate Suggest
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/cocorof-whiteboard-curate-suggest/badge)](https://www.skillsdirectory.com/skills/cocorof-whiteboard-curate-suggest)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: whiteboard-curate-suggest
description: When an insight, decision, or checklist with lasting value comes up during a conversation, suggest saving it to the library, and on the user's consent register it in Curated Knowledge via knowledge_promote.
allowed_tools:
  - knowledge_promote
  - opsidian_read
  - opsidian_search
execution_mode: inline
examples:
  - "Should I save what we just organized to the library?"
  - "Save this for me."
  - "This seems worth reusing next time."
---

# Whiteboard: Curate Suggest — Suggest saving content with lasting value

When content with lasting value comes up during a conversation, suggest **saving it to the library** to the user, or run `knowledge_promote` when the user explicitly asks.

## Understanding the two vaults

- **User Opsidian** — the user's raw personal vault. Notes that accumulate daily.
- **Curated Knowledge** (library) — the subset of *lasting value* the user has hand-picked themselves. You (the VTuber) can freely search this via `knowledge_search`. The more organized notes pile up in the library, **the smarter your future turns get too**.

So a promote suggestion is *mutually beneficial* — don't suggest it too often, nor too rarely.

## When to *suggest*

Suggest once when **two or more** of these signals appear at the same time:

- **A conclusive utterance**: "X is the answer after all", "let's go with this", "I'll remember that", "that wraps it up"
- After a **5+ turn discussion on one topic**, an agreed-upon key point emerges
- A **reusable structure** — a procedure / numbers / a checklist / a set of commands / a decision tree
- **Valuable info brought in from outside** — a tool / technique / link the user jotted down as if seeing it for the first time

**Signals NOT to suggest on**:
- Chit-chat, greetings, expressions of emotion
- One-off questions (today's weather, a simple info lookup)
- Utterances from the user that are negative / private / contain personal information
- A note clearly likely to already be in the library (a result the user just found via `whiteboard-search`)

## How to *suggest*

Keep it concise:

> "Want me to keep what we've organized so far in the library? It'll be faster to find next time."

Or:

> "I think it'd be good to save this — should I promote it?"

**Forbidden formats**:
- Prompting at the end of every turn (don't prompt twice on the same topic within one session)
- Explaining cost / limits / technical terms along with the prompt (don't expose metadata like "the vector indexing is progressing…")

## Handling the user's response

- **Explicit consent** ("yeah", "sure", "yep", "save it", "save it") → go to the execution steps in the next section.
- **No response / switching to another topic** → stay silent. **Don't re-prompt** (this session, on the same topic only).
- **Explicit refusal** ("nope", "I'm good") → briefly acknowledge, e.g. "got it — let me know if you need it later".

## How to *execute* (after the user consents)

1. **Decide the source note**:
   - If an existing note in User Opsidian is connected to this conversation, use that note.
     - If unsure, do a quick **`opsidian_search(query="...")`** or briefly check with the user: "which note should I base it on?".
   - If the conversation itself has only new content and there's no source note:
     - Guide the user, e.g. "then I'll organize it and capture one line in the inbox — can you give me a one-line summary?".
     - **Don't create a new note on your own arbitrarily** — having the user confirm once is the heart of trust.
2. Call **`knowledge_promote(filename="...")`**.
3. Check the response → confirm to the user:

> "I saved it to the library as '[title]'. Later, if you say 'show me the library stuff about X', I'll find it right away."

- Don't expose the raw `curated_filename` — title only.
- On failure (already promoted, permissions, etc.) → paraphrase the reason for the user in one natural-language line.

## Example responses

**Good example (suggest → consent → execute)**:

User: "Yeah, in the end the answer for retry is exponential backoff + jitter"
You (suggest): "Want me to keep this in the library? Next time you can pull it up right away in a similar case."
User: "Yeah, ok"
You: *(find the source with opsidian_search('retry exponential'))*
*(knowledge_promote(filename='topics/retry-patterns.md'))*
"I saved it to the library as 'Retry patterns — backoff + jitter'. Next time you ask 'how do I do retry?' I'll pull it right up."

**Good example (refusal)**:

You: "Want me to save this?"
User: "No, just talking for now"
You: "Got it — let me know if you need it later."

**Bad examples (never do this)**:

❌ Prompting "save it?" every turn
❌ Auto-calling `knowledge_promote` without the user's consent
❌ Exposing internal phrasing like "registered in the Curated FAISS vector index"
❌ Trying to promote the same note twice in the same session

## Don'ts

- **Absolutely never call promote without the user's consent.**
- No prompting every turn — if the "when to suggest" conditions above aren't met, stay silent.
- Never prompt for content where promoting is inappropriate (personal info, one-off chit-chat).
- Don't try to promote the same note twice in the same session (the tool response may contain a hint).
- Don't expose system terms like raw filename / capture_id / FAISS / vector.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.