Skip to main content
skill CX Codex

archive-conversation — Claude Code skill: archive a chat session

The archive-conversation skill turns an AI conversation into an archival note: what was decided, how the thinking moved, what stayed open. Useful when a session has to outlive the tab.

MIT 117 installs ★ 10 updated 2026-10-19

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add nweii/agent-stuff --skill archive-conversation -a claude-code
Open on GitHub

What archive-conversation does

It reads the whole conversation before writing anything, then builds a note for someone who comes back without the transcript. Decisions and the reasoning behind them, the drift of the discussion, the user's own wording, loose ends — a reader should be able to recover all of it from the note alone. The shape follows the talk: a debugging thread becomes what broke, which attempts failed, the fix; a work session becomes the goal, the actions and files touched, the current state.

Statements that carry a decision, a doubt or a moment of recognition get quoted word for word, with only typos cleaned up, while the assistant's replies stay in its own voice. Section headings are sentence case and describe what happened inside them.

Default filename is {{Type}} - {{topic}} YYYY-MM.md, with Thinking for insight-heavy talks and Log for work sessions. A path passed as [save-location] wins; otherwise the skill reads two or three of your similar notes and copies their naming and filing. With no file access it prints the note as a Markdown code block under the intended path, so nothing reaches the disk.

What it helps with

  • keep a chat session as a note you can reread later
  • record why a decision was made, not just what it was
  • preserve the user's exact wording in the archive
  • log files changed and tasks left over after a work session
  • match new notes to an existing archive folder

Example prompts for the agent

  • Archive this conversation to notes/archive/
  • Write up this session next to my other archive notes
  • Save an archival note for this debugging chat

How to install archive-conversation

1Skills CLI (Vercel)

npx skills add nweii/agent-stuff --skill archive-conversation -a claude-code

2Manually: put the files in place

~/.claude/skills/archive-conversation/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/archive-conversation/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about archive-conversation

How do I install archive-conversation in Claude Code?

Run npx skills add nweii/agent-stuff --skill archive-conversation -a claude-code. The skill lands at ~/.claude/skills/archive-conversation/SKILL.md.

What does archive-conversation do?

It writes an archival note about an AI conversation: decisions with their reasoning, how the thinking developed, verbatim quotes from the user, files touched, open questions.

Where does it save the note?

Pass a path as [save-location] and it goes there. Without one, the skill looks at your similar notes and files the new one alongside them; with no file access it prints a Markdown block instead.

Source text

SKILL.md · MIT
---
name: archive-conversation
description: "Create analytical archival summaries of AI conversations, capturing intellectual journeys, key insights, and technical logs. Use when archiving, saving, or documenting a chat session."
argument-hint: "[save-location]"
metadata:
  author: nweii
  version: "1.3.0"
---

# AI Conversation Archival Summary

Create an archival summary of an AI conversation that captures its intellectual journey, key insights, or technical work session logs. Document either how thinking evolved throughout the discussion or the specific actions and technical decisions made during a work session.

## What the Note Is For

The note is a record for a reader coming back without the conversation at hand. A good note lets that reader recover:

- what was decided, and why
- how the thinking moved through the back-and-forth
- the user's own thinking, in their words
- what was produced and where it lives
- what stayed open

The note is done when everything on that list the conversation contains is recoverable from the note alone.

## Analyze the Whole Conversation

Read the entire conversation before writing. Name its nature (work session, creative exploration, learning, strategy) and what made it worth saving. Look for:

- the real question beneath the first one
- where assumptions changed or decisions were made
- frameworks or patterns that emerged along the way
- dead ends that taught something
- connections between ideas or system parts
- the human texture: confusion, recognition, frustration, breakthroughs

## Shape the Note

Let the conversation's flow set the structure. Common shapes:

- **Problem-solving:** what broke → instructive failed approaches → the working solution → next steps
- **Creative exploration:** the initial vision → how ideas branched → key decisions and why → directions worth revisiting
- **Learning:** what wasn't understood → how understanding built → breakthroughs → remaining questions
- **Work session:** the objective → actions taken and files changed → hurdles and fixes → current state and remaining tasks
- **Strategy:** the decision → options and trade-offs → criteria that emerged → next considerations

Give each section a sentence-case heading that says what happened in it, such as "Starting from hourly vs. project pricing questions" or "Why the recursive function kept hitting memory limits." Write in plain, third-person documentation style; first person appears only in the user's quoted words.

## Preserve the User's Own Words

Quote the user **verbatim** wherever they state a decision, reasoning, opinion, worry, excitement, or an idea they want to pursue. Their words record their thinking at the time, close to a journal entry; a paraphrase flattens their voice and the back-and-forth in which ideas took shape.

- Open the section a quote informs with the quote; analysis follows.
- Keep their wording and casing; correct only typos and dictation errors.
- Quote whole thoughts.
- Keep the AI's contributions in the AI's voice.
- Be selective about which moments, not about how much of each.

When an exchange shows thinking changing, quote both sides:

> [User's first name, if known]: "[moment of recognition, doubt, or decision]"
> AI: "[response that shifted understanding or articulated a key insight]"

## File Output Requirements

Match the user's system first. Use what you already know of their conventions from their instructions, memory, or this conversation. Otherwise, before writing, read two or three of their most similar notes (earlier archives, or recent notes in the likely folder) and follow how those are named, filed, and tagged. Use the defaults below only when there is no system to match.

### Naming

Default format: `{{Type}} - {{topic}} YYYY-MM.md`, with `Thinking` for insight-heavy conversations and `Log` for work sessions.

- Example: `Thinking - Portfolio strategy 2025-08.md`
- Example: `Log - Refactoring auth middleware 2025-01.md`

### Save Location

1. **Save-location argument:** use that path directly.
2. **Existing structure:** save next to the user's similar notes, or in the folder that best fits the conversation's subject.
3. **Unclear:** ask the user where to save.
4. **No file access:** output the note as a Markdown code block with the intended path above it.

### Metadata

Fill only the frontmatter fields those similar notes share and this conversation can answer accurately, and reuse tags that already exist. With no system to match, add the date and a few tags for the conversation's topic.

© nweii · MIT · nweii/agent-stuff

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set2

Project sponsors

Companies that keep this analytics open