Health checks

A brain is a living thing. It can be healthy or it can be quietly rotting. The difference between a brain that helps you in year three and one that you abandon at month six is a small set of routine checks that take very little time.

This file lists what to run, when to run it, and what the signals mean.

Weekly: run the linter

Once a week, open your AI chat and ask it to run the lint-check skill. A simple prompt is enough:

Run the lint-check skill on this brain. Report orphan pages, broken wikilinks, missing frontmatter, and anything in raw/ older than 30 days.

What you get back is a list. It will look something like:

• 3 orphan pages in wiki/concepts/ with no incoming wikilinks
• 1 broken wikilink: wiki/projects/launch-coffee-shop.md points to [[jane-d]] which does not exist
• 7 notes missing frontmatter (acceptable, just FYI)
• 2 files in raw/ from more than 30 days ago

Fix the broken wikilinks first. A broken link is a thought you wanted to connect to something that is not there yet. Either create the missing note or remove the link. Both are fine.

Orphan pages are pages with no incoming links. Not always a problem, sometimes the page genuinely does stand alone. But a cluster of orphans usually means a topic that should be connected to the rest of the brain and is not. Click into each one and add two or three wikilinks.

Missing frontmatter is a soft warning. Fix it if you care, ignore it if you do not.

Files lingering in raw/ are sources you forgot to ingest, or originals you forgot to move out after ingest. Either ingest them now or move them to your long-term storage and delete the raw copy.

Monthly: refresh the knowledge graph

Once a month, ask the chat to run a full graphify pass:

Run graphify on this brain. Produce an updated knowledge graph in graphify-out/. Highlight changes from the previous run.

Graphify reads everything in raw/ and wiki/, extracts entities, builds the connections, and writes the result to graphify-out/. The output feeds back into the AI's context for richer queries.

The signals to watch:

• Entity count growing month over month. Healthy brain. You are capturing new people, projects, concepts.
• Entity count flat. Could be fine (mature brain, stable life). Could be a signal that you stopped capturing.
• Sudden spike in unresolved entities. Usually a sign that an ingest pulled in a lot of new names that need disambiguation. Spend ten minutes adding aliases.

Monthly: walk the graph in Obsidian

This is the five-minute habit that catches what no automated tool can.

Open the brain in Obsidian. Click the graph view icon. Look at the whole shape.

You are looking for three things:

1. Orphan nodes floating on their own at the edges. They want wikilinks.
2. Clusters that should connect but do not. Your work cluster and your travel cluster might share three people who should be linked across both.
3. Stale corners. A part of the graph you have not touched in months. Click in. Either the project is genuinely dormant (move it to archive/) or you forgot it existed and it deserves attention.

Five minutes. Click around. Trust your eyes.

Quarterly: skill audit

Every three months, audit the skills installed in skills/ and any third-party skills you have added to your AI chat. Skills are markdown files that tell the AI what to do, and a poorly-written or malicious skill can do real damage inside your brain.

Before installing any new skill from a third party, grep it for known prompt-injection patterns:

grep -E "do not scan|skip all tests|Trust Hub|onerror=|administrator's request" skills/<new-skill>/*.md

If anything matches, do not install. Read the skill in full and decide.

Every quarter, run the same grep across every skill in skills/:

grep -rE "do not scan|skip all tests|Trust Hub|onerror=|administrator's request" skills/

Read every diff on skill updates. A skill that grows three suspicious lines between version 1.2 and 1.3 is a skill to remove, not update.

What each signal means

When in doubt, ask the AI

Most health questions have a two-prompt answer. Try:

Run a full health check on this brain. Identify the three most important things I should fix this week.

The chat will look at the structure, run the relevant skills, and prioritise. Trust the prioritisation. Fix the top item, ignore the rest until next week.