home / skills / simhacker / moollm / summarize

summarize skill

safe

/skills/summarize

This skill summarizes long materials while preserving key truths and backlinks to sources for reliable reuse.

npx playbooks add skill simhacker/moollm --skill summarize

Review the files below or copy the command above to add this skill to your agents.

Files (6)

SKILL.md

2.7 KB

---
name: summarize
description: "Compress without losing truth. Backlink to sources."
license: MIT
tier: 1
allowed-tools:
  - read_file
  - write_file
related: [honest-forget, session-log, memory-palace, play-learn-lift, research-notebook]
tags: [moollm, compression, knowledge, context, backlinks]
---

# Summarize

> **Compress without losing truth. Backlink to sources.**

Context compression for memory management.

> [!IMPORTANT]
> **Always backlink.** Every summary points to its source. Never orphan knowledge.

## The Goal

When files are too large for context:
1. **Summarize** — Extract key information
2. **Backlink** — Point to original source
3. **Prioritize** — Hot/cold hints for what matters
4. **Preserve** — Never delete, just compress

## Contents

| File | Purpose |
|------|---------|
| [SKILL.md](./SKILL.md) | Full protocol documentation |
| [SUMMARIES.yml.tmpl](./SUMMARIES.yml.tmpl) | Summary template |

## Example

```yaml
summary:
  source: "designs/original-design.md"
  created: "2025-12-30"
  
  key_points:
    - "Files are state, no hidden memory"
    - "YAML comments carry meaning"
    
  backlink: "../designs/original-design.md"
  full_context_needed_for:
    - "Implementation details"
    - "Edge cases"
```

## The Intertwingularity

Summarize enables the LIFT stage — compress wisdom for sharing.

```mermaid
graph LR
    SUM[📝 summarize] -->|compresses| SL[📜 session-log]
    SUM -->|compresses| RN[📓 research-notebook]
    SUM -->|enables| HF[🌫️ honest-forget]
    SUM -->|part of| PLL[🎮📚🚀 play-learn-lift LIFT]
    
    SR[🔧 self-repair] -->|triggers| SUM
```

---

## Dovetails With

### Sister Skills
| Skill | Relationship |
|-------|--------------|
| [play-learn-lift/](../play-learn-lift/) | Summarize IS LIFT — share wisdom |
| [honest-forget/](../honest-forget/) | Summarize before forgetting |
| [session-log/](../session-log/) | Source material to compress |
| [research-notebook/](../research-notebook/) | Findings to distill |
| [self-repair/](../self-repair/) | Triggers when context exceeds budget |
| [memory-palace/](../memory-palace/) | Place summaries in palace rooms |

### Protocol Symbols
| Symbol | Link |
|--------|------|
| `SUMMARIZE` | [PROTOCOLS.yml](../../PROTOCOLS.yml#SUMMARIZE) |
| `HONEST-FORGET` | [PROTOCOLS.yml](../../PROTOCOLS.yml#HONEST-FORGET) |
| `HOT-COLD` | [PROTOCOLS.yml](../../PROTOCOLS.yml#HOT-COLD) |

### Kernel
- [kernel/memory-management-protocol.md](../../kernel/memory-management-protocol.md) — Full specification

### Navigation
| Direction | Destination |
|-----------|-------------|
| ⬆️ Up | [skills/](../) |
| ⬆️⬆️ Root | [Project Root](../../) |
| 🌫️ Sister | [honest-forget/](../honest-forget/) |
| 🎮 Sister | [play-learn-lift/](../play-learn-lift/) |

Overview

This skill compresses source documents into faithful, concise summaries while always providing a backlink to the original. It is designed to reduce context size for memory-limited systems without discarding truth or provenance. Summaries include priority hints (hot/cold) and notes about when full context is required.

How this skill works

The skill extracts key points, metadata, and explicit backlinks for every source so no summary becomes orphaned. It tags which topics or sections are 'hot' (likely needed soon) versus 'cold' (safe to deprioritize) and records what full-context items must be preserved for future implementation or edge-case review. Outputs follow a stable template: source, created timestamp, key_points, backlink, and full_context_needed_for.

When to use it

When a file or session exceeds available context budget and must be compressed.
Before performing an honest forget operation to avoid data loss.
When preparing notes for long-term memory or a memory palace.
When sharing distilled findings across teams or agents with provenance.
When you need prioritized hints about what to load back at retrieval time.

Best practices

Always include an explicit, working backlink to the exact original source.
Keep summaries factual and minimal: preserve claims, omit speculation.
Mark full-context requirements clearly so implementers know what to restore.
Use hot/cold tags to guide future retrieval and prioritize memory budget.
Treat files as state: never delete original content—compress, don’t discard.

Example use cases

Compressing a long research note into a one-paragraph summary with backlink and a list of experiments requiring full context.
Summarizing a session log after a debugging session, marking error reproductions as hot and alternative ideas as cold.
Preparing design briefs for cross-team review that link back to full specifications for implementation details.
Running a memory-management pass that compresses old messages while preserving provenance and restore hints.

FAQ

Will the summary ever omit the original source link?

No. Every summary must include a backlink to the original source so the full context can be retrieved.

How do hot/cold hints affect retrieval?

Hot items are prioritized for rehydration into context; cold items remain compressed until explicitly requested or when budget permits.