Review our entire conversation from the beginning. Your goal is to distill a reusable learned skill that captures the workflow pattern, failure modes, and key corrections.

Analyze the following aspects:

1. **Workflow Pattern**: What tools were used and in what sequence? What was the overall approach?
2. **Failure Modes**: Where did things go wrong? What errors occurred? How were they recovered from?
3. **User Corrections**: Where did the user steer the approach differently? What did they correct or improve?
4. **Key Parameters**: What specific configurations, thresholds, or settings were important?
5. **Reproducibility**: What would someone need to know to reproduce this workflow on a similar problem?

Then call the `learn` tool with:
- `name`: A descriptive kebab-case identifier (e.g., "debug-cuda-memory-leak", "deploy-vercel-prebuilt")
- `description`: A one-line summary of what this skill teaches
- `content`: A complete SKILL.md with this structure:

```
---
name: {name}
description: {description}
source: rsi
metadata:
    skill-author: /learn command
---

# {name}

## Overview
[What this skill covers and when to use it]

## Workflow Pattern
[Numbered steps with tool names and key actions]

## Failure Modes & Recovery
[What went wrong and how it was fixed — the most valuable part]

## User Corrections
[Where the user steered the approach — captures human judgment]

## Key Parameters
[Important configurations, thresholds, file paths]

## When to Use
[Trigger conditions — what kind of problem/question matches this skill]
```

$ARGUMENTS
