Save Conversation

Capture the current conversation as a markdown file in a standardized folder structure. Unlike automatic session logs, this is selective - only save conversations worth preserving.

---

When to Use

• Completed a meaningful task you want to reference later
• Had a discussion with important decisions or insights
• Debugging session with a solution worth documenting
• Research conversation with findings to preserve
• Any conversation that shouldn't disappear into the void

---

Configuration

Storage Location

Default: saved-conversations/ in the current working directory.

Create a .conversation-config file to customize:

storage_path: ./my-conversations

Or use environment variable: CONVERSATION_STORAGE_PATH

---

Workflow

Step 1: Determine Topic Name

If provided as argument ($ARGUMENTS), use it.

Otherwise, derive from the conversation:

• Identify the primary goal or topic discussed
• Use lowercase-with-hyphens format
• Be descriptive: debug-auth-token-refresh not debugging
• Prefix with category when appropriate: research-, fix-, setup-

If unclear, ask the user for a topic name.

Step 2: Check Storage Location

Check for .conversation-config file:

cat .conversation-config 2>/dev/null || echo "not found"

Check environment variable:

echo "${CONVERSATION_STORAGE_PATH:-not set}"

Default to saved-conversations/ if neither exists.

Step 3: Generate Filename

Format: YYYY-MM-DD_[topic-name].md

Example: 2025-02-20_debug-auth-token-refresh.md

If file already exists, append incrementing number:

• 2025-02-20_topic.md
• 2025-02-20_topic-2.md
• 2025-02-20_topic-3.md

Step 4: Analyze Conversation

Review the current conversation to extract:

1. Context: What triggered this conversation? What was the starting point?
2. Goal: What was the user trying to accomplish?
3. Key Exchanges: The important back-and-forth (not every message)
4. Decisions Made: Choices and their rationale
5. Actions Taken: What was actually done (files changed, commands run)
6. Outcome: What was the result? Successful? Partial? Needs follow-up?
7. Insights: Learnings, gotchas, or things to remember

Step 5: Generate Markdown

Structure the file as:

# [Topic Title]

**Date:** YYYY-MM-DD HH:MM
**Duration:** [Approximate conversation length]
**Outcome:** [Success | Partial | Needs Follow-up | Research]

## Summary

[2-4 sentence overview of what happened]

## Context

[What triggered this conversation? What was the starting state?]

## Goal

[What was the user trying to accomplish?]

## Key Exchanges

### [Exchange 1 Topic]

**User:** [Summarized request or question]

**Claude:** [Summarized response or action]

[Repeat for other significant exchanges]

## Decisions Made

- **[Decision]:** [Rationale]
- **[Decision]:** [Rationale]

## Actions Taken

- [Action 1] → [Result]
- [Action 2] → [Result]

## Outcome

[What was accomplished? What's the final state?]

## Insights

- [Insight or learning 1]
- [Insight or learning 2]

## Follow-up

- [ ] [Any remaining tasks or things to revisit]

---

*Saved from conversation on YYYY-MM-DD*

Step 6: Create Directory and Save

mkdir -p [storage_path]

Write the markdown file:

# Use Write tool to save the file

Step 7: Confirm to User

Report:

Conversation saved to: [full path]

Summary:
- Topic: [topic name]
- Date: [date]
- Outcome: [outcome status]
- [N] key exchanges captured
- [N] actions documented

---

Extraction Guidelines

What to Include

• Key decisions: Choices with significant impact
• Important commands/code: Working solutions, not debugging attempts
• Insights: Things learned that weren't obvious
• Final state: What exists now after the conversation

What to Exclude

• Trivial exchanges ("Let me check that", "Sure")
• Failed attempts and debugging dead-ends (unless they contain lessons)
• Repeated/redundant information
• Overly detailed step-by-step (summarize instead)

Summarization Level

Conversation Type	Detail Level
Quick fix/task	Brief - just outcome and solution
Complex debugging	Medium - include key discoveries
Research/exploration	Detailed - capture findings
Decision-making	Focus on options considered and rationale

---

Examples

Example 1: Debugging Session

User: /save-conversation debug-api-timeout

Extracted:
- Context: API calls failing with 504 errors
- Goal: Fix timeout issues in production
- Key exchange: Identified retry logic bug
- Decision: Increased timeout to 30s, added exponential backoff
- Actions: Modified api-client.ts, deployed to staging
- Outcome: Success - errors resolved
- Insight: Default axios timeout was too low for our backend

Example 2: Research Discussion

User: /save-conversation

Claude: What topic should I use for this conversation?

User: comparing-auth-libraries

Extracted:
- Context: Evaluating auth options for new project
- Goal: Choose between Auth0, Clerk, and custom JWT
- Key exchanges: Pros/cons discussion, pricing analysis
- Decision: Go with Clerk for this project
- Actions: None (research only)
- Outcome: Research complete
- Insights: Clerk has better DX, Auth0 better enterprise features

---

Tips

• Save meaningful conversations only: This is selective archival, not automatic logging
• Name descriptively: Future you should understand the topic from the filename
• Save before context fades: Best to save right after concluding
• Edit after saving: The generated file is a starting point - refine if needed

---

Related Skills

Skill	Purpose
save-playbook	Save as reusable workflow, not just record
json-memory:update-memory	Update persistent agent memory
brain-memory:create-note	Create atomic knowledge note