Save Answer

Save Answer

Files a conversation output back into the brain vault. Turns ephemeral Q&A, research, analysis, and comparisons into persistent, cross-linked knowledge.

Inspired by Karpathy's LLM Wiki pattern: "good answers should be filed back into the wiki as new pages so explorations compound."

Trigger phrases

• /save-answer
• "save this answer"
• "file this"
• "keep this in the brain"
• "this is worth saving"
• "persist this"

Input

The content to save. This can be:

• Explicit: User says "save this" after a Q&A exchange or analysis
• Implicit: User invokes /save-answer and Claude identifies the most recent substantial output in the conversation

If no clear output is identifiable, ask: "Which part of our conversation should I save?"

Step-by-step workflow

1. Identify the output

Extract the valuable content from the conversation. Common types:

• Research finding -- answer to a question that required multi-source synthesis
• Comparison/analysis -- a table or structured evaluation of options
• Framework application -- a mental model applied to a specific problem
• Decision rationale -- reasoning behind a choice (if not already logged as a decision)
• Technical insight -- something learned about a tool, API, or system

2. Determine placement

| Type | Location | |------|----------| | Project-specific insight | Append to projects/{project}/{relevant-file}.md | | Research finding | research/{topic}/{descriptive-slug}.md | | Cross-project analysis | research/{topic}/{descriptive-slug}.md | | Decision rationale | decisions/YYYY-MM-DD-{slug}.md (use decision template) |

Check for existing notes on the same topic before creating new files. Update if a close match exists.

3. Write the content

For new notes:

---
title: "{Descriptive title}"
date: YYYY-MM-DD
type: note
tags: [{relevant, tags}]
source: "conversation"
---

# {Descriptive title}

> Derived from conversation on YYYY-MM-DD

## Context

{1-2 sentences: what question or analysis led to this}

## Key Findings

{The actual content -- distilled, not a transcript dump:
- Lead with conclusions
- Include specific data, comparisons, or frameworks
- Keep it scannable}

## Relevance

- [[project-or-note]] -- {how it connects}

For appending to existing notes:

• Add under the most relevant section (## Notes, ## Research, or a new subsection)
• Include a date marker: *Added YYYY-MM-DD from conversation*
• Keep it concise

4. Cross-link

• Add [[wikilinks]] to related notes in the saved content
• If highly relevant to a project, add a backlink in the project note

5. Append to activity log

Add an entry to log.md:

## [YYYY-MM-DD] save | {Short title}
- **Type:** {research finding | comparison | framework | decision rationale | technical insight}
- **Action:** {Created | Updated} `{file path}`
- **Linked to:** [[note-a]], [[note-b]]

6. Report to user

Saved: "{title}"
Location: {file path}
Action: {Created new note | Appended to existing note}
Linked to: [[note-a]], [[note-b]]

Content rules

• Distill, don't transcribe. Extract the insight, not the conversation.
• Only save durable knowledge. Skip anything that won't be useful in 3+ months.
• One concept per note. If the output covers multiple distinct topics, split into separate notes.
• Don't duplicate. Check research/ and projects/ before creating.
• Don't touch curated collections. Never merge into knowledge/lenny/ or other hand-curated files.

Proactive use

The /wrap-session-up skill checks for unsaved session outputs in its "Compound session outputs" step. This skill can be invoked directly when the user wants to save something mid-conversation without waiting for wrap-up.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: michaeljauk
• Source: michaeljauk/brain-starter
• License: MIT

Install and usage instructions live in the source repository linked above.