Skip to content
Back to skills

Memory Connections

ASecurity

Show a memory's connections to other memories, grouped by relationship (parent, child, sibling). Pass the memory's ID or title in the argument.

  • 2 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 23, 2026
ai-agents

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add oxHive/mynd --skill memory-connections --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Memory Connections?

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

Security grade badge for Memory Connections
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/oxhive-memory-connections/badge)](https://www.skillsdirectory.com/skills/oxhive-memory-connections)

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: memory-connections
description: Show a memory's connections to other memories, grouped by relationship (parent, child, sibling). Pass the memory's ID or title in the argument.
---

Show how a memory connects to the rest of what's stored in Mynd.

Input from $ARGUMENTS: a memory ID or title to look up.

## Process

### 1. Resolve the memory

- If $ARGUMENTS looks like a memory ID (starts with "mem_"), call `memory_recall` with `id`.
- Otherwise call `memory_recall` with `title`.
- If not found, say so and suggest using `/memory-search` to find the right ID.

### 2. Get its connections

Call `memory_get_edges` with the resolved memory's ID.

### 3. Present grouped

Show three sections, always all three even when empty:
- **Parents** (broader context this falls under) — title + id, and the link phrase if present.
- **Children** (specific instances of this) — same format.
- **Siblings** (peers, no hierarchy) — same format.

If a section has no entries, say "(none)" rather than omitting the heading.

A memory's children may be **fragments of one logical document** rather than independent
topics — this is the index/fragment chunking convention (see `/memory-store`'s chunking
step). Don't guess this from phrasing or tag similarity: check the `part` tag directly.
`part:index` on the resolved memory means its children (or at least the ones also tagged
`part:fragment`) are pieces of one document, not separate topics. When that's the case,
say so when presenting the children — e.g. "Children (fragments of this document):" — so
the user understands why several "children" are pieces of one thing, not several unrelated
topics. If the resolved memory has no `part` tag, its children are independent topics as
normal — don't apply the fragment framing speculatively.

## Rules

- Never fabricate connections. Only show what `memory_get_edges` returned.
- If the user wants to explore a connected memory further, offer to `memory_recall` it or run `/memory-connections` on it next.
- A memory's content may contain inline mention links like `[label](mem_xxx)` pointing at
  a parent/child/sibling. Treat these as lazy pointers, not required reading: follow one
  only when the current task actually needs that related memory's content. Don't
  proactively `memory_recall` every mentioned id just because it's linked. This applies
  doubly to `part:fragment` children — recalling every fragment of a document just to
  answer a question about one section defeats the purpose of chunking in the first place.

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.