Your Best Thinking Is Dying in DMs

Count how many times you have typed the same advice this month. The pricing objection script. The "here's how I'd structure that seed round" reply. The three questions you always ask before someone hires their first salesperson. If you are good at what you do, the number is embarrassing, and every one of those messages is sitting in a Slack thread or a WhatsApp chat that nobody, including you, will ever read again.

That is the asset. Not your deck, not your blog, not your conference talk. The advice you repeat under mild time pressure to people you actually want to help is the most distilled version of your thinking that exists. And its distribution channel is a medium with the retention of a napkin.

DMs are where thinking goes to die

A DM has three properties that make it terrible at preserving knowledge:

1. Audience of one. You spent fifteen minutes writing a genuinely good answer and exactly one person benefits. The marginal cost of helping the next person is the full fifteen minutes again.
2. Zero searchability. Even you can't find it. "I know I wrote this exact thing to someone in March" is a sentence every operator has said, followed by rewriting it from scratch, worse.
3. No compounding. A blog post gets refined over drafts. Your DM advice gets rewritten from zero every time, so it never improves. Version 40 of the answer is no better than version 3, it's just typed faster and with more typos.

The cruel part is that repetition is the signal. If you've explained your hiring bar four times this quarter, that explanation is validated content. People pulled it out of you. You never had to guess whether anyone wanted it, the way you do with a blog post.

The blog post is not the fix

The standard advice is "turn it into a blog post." Fine, do that too. But a post is static and generic, and the DM was neither. When a founder asks you about churn, you don't send them your churn essay. You ask what their contract length is, whether it's logo or revenue churn, who owns the renewal, and then you give a specific answer shaped by your actual experience. The value was in the application, not the artifact.

What you actually want is the function, not the output. Something that takes a new situation as input and produces your answer, the way you would give it, with your priors and your scar tissue. Until recently that function only ran on your own wetware, one conversation at a time.

Capture it once, as a skill

Claude Code plugins have four surfaces: skills, commands, hooks, and MCP servers. For personal expertise, the one that matters is the skill. A skill is just a directory with a SKILL.md in it:

advising-on-churn/
├── SKILL.md          # frontmatter: name, description; body: how you think
└── references/
    ├── frameworks.md # your actual mental models, written down once
    └── examples.md   # real (anonymized) cases you've worked through

The SKILL.md has YAML frontmatter (name, description) and then plain Markdown describing how you reason about the problem. The description is what tells Claude when to load it; skills are loaded on demand, so you can pack in real depth without burning context on every conversation. The references/ directory is where the fifteen-minute DM answers go. Every framework you've retyped, written down one final time.

The difference between this and a notes app is that a skill executes. Someone doesn't read your churn framework, they ask Claude a churn question and get your framework applied to their numbers, in your voice, with your standard follow-up questions asked first. It's the DM conversation, minus you having to be awake for it.

Distribution is two commands

A skill on your laptop helps only you. Packaging it as a plugin makes it installable: a .claude-plugin/ directory with a plugin.json, listed in a marketplace.json that Claude Code can add by URL. One sharp edge if you host this yourself: plugin sources served over a URL must be git-backed (the git-subdir source type). Point a marketplace entry at a relative HTTP path and the install fails silently, which is a miserable hour of debugging the first time.

Once it's hosted, anyone with Claude Code runs:

/plugin marketplace add https://yourdomain.com/you
/plugin install you@your-marketplace

Two commands, verified on Claude Code CLI 2.1.x, and your accumulated pattern-matching is now a tool inside someone else's terminal. The founder who would have DMed you at 11pm asks the plugin instead. They get a better answer than your tired thumbs would have produced, and you get your evening back.

What installs.me automates

You could write all these SKILL.md files by hand. Most people won't, for the same reason the blog posts never got written: the knowledge is in your head and your message history, not in a document, and extraction is the expensive part.

installs.me does the extraction. You connect the places your thinking already lives, your files, your Google Drive, your calendar, and Claude synthesizes the persona plugin from them: the skills, the reference files, the voice. The advice you've been retyping weekly becomes the training material. The output is a hosted marketplace anyone can add.

The demo is a real person. Lautaro Schiaffino built and sold Sirena for $30M, now runs Darwin AI, angel invests across LatAm, and has answered the same founder questions hundreds of times in WhatsApp. His persona is installable right now, which means his 11pm DM answers are available at any hour, to anyone, without him typing a single one of them again.

Your best thinking deserves better than a chat scrollback. Write it down once, or better, let the machine write it down from what you've already produced, and then let it run without you.

Install a person

installs.me turns your files, calendar and calls into a Claude Code plugin that thinks like you. Anyone installs it with two commands:

/plugin marketplace add https://installs.me/lautaro
/plugin install lautaro@lautaro-installs