Run one prompt against your own usage history and it tells you exactly where to invest your automation effort.
The Prompt
Ask Claude to scrape all of your sessions on a machine and produce a breakdown of:
- Everything you do repeatedly
- What should become a skill (slash command)
- What should become a plugin (MCP server)
- What should become an agent (autonomous subprocess)
- What should live in CLAUDE.md (persistent instructions)
This single prompt is a complete workflow audit. It turns your accumulated usage into an optimization roadmap.
Why This Works
Most people never audit their own tooling habits. They add skills and shortcuts reactively, one at a time, when friction becomes painful enough. The session audit inverts this. It gives you a top-down view of your actual behavior, not what you think you do, but what you actually do.
Your Claude session history is a near-perfect log of your real workflows. Every question you asked, every file you edited, every command you ran. It is the ground truth of how you operate. Running analysis against it reveals:
- Repeated patterns you stopped noticing. The thing you do 5 times a week that takes 30 seconds each time. Individually trivial. Collectively, a skill waiting to be extracted.
- Classification boundaries. Not everything should be the same type of automation. The audit forces you to think about the right abstraction level for each pattern.
- Gaps in your CLAUDE.md. Instructions you keep giving verbally that should be written down permanently.
- Agent opportunities. Multi-step workflows you run manually that could be fully autonomous.
The Classification Framework
| Type | When to use | Example |
|---|---|---|
| CLAUDE.md | Persistent context, preferences, conventions | “Always use 1Password CLI for secrets” |
| Skill | Repeatable single-action workflows triggered by the user | /commit, /review-pr, a custom /deploy |
| MCP Plugin | External data sources or services Claude needs to query | Semantic search over a knowledge base, Slack integration |
| Agent | Multi-step autonomous tasks that run independently | “Scrape all sessions and classify patterns” |
The key distinction: skills are user-triggered and single-purpose. Agents are autonomous and multi-step. Plugins connect to external systems. CLAUDE.md is always-on context.
How to Run It
Scrape all of my Claude sessions on this computer. Give me a breakdown of
all the things I do, things that are worth making into skills vs plugins
vs agents vs CLAUDE.md instructions.
Then act on the output. Build the top 3 highest-leverage automations first. Re-run the audit monthly.
This Is Zero-Cost Knowledge Extraction Applied to Yourself
This is zero-cost knowledge extraction turned inward. Your own session history is a high-signal source. The extraction cost is near zero. The value is a complete map of where your time goes and how to reclaim it.
The person who audits their own workflows and automates the top patterns compounds faster than the person who just uses the tools as-is.
