Day 9 of 23 · Monday · Learning

Operating the machine vs building itReview

Day 9 closes the catch-up block. Today isn't a new mechanism — it's the shift from building more machinery to steering and pruning what you already run. The hands-on is an honest audit of your own skills list.

Catch-up progress

9/23

Why this matters to you

Your workspace has quietly accreted dozens of skills — gstack, superpowers, the plaud-* bundle, document-skills, claude-mem, a dozen first-party ones. Your own goals.md lists 'Optimizing tools that already work' under Things NOT Worth My Time. The real risk on a build-heavy sabbatical isn't too few capabilities — it's accreting them faster than you can actually wield them, until the menu itself becomes the bottleneck.

There are two different jobs hiding inside 'working on your system.' Building is adding a new lever — a new skill, a new LaunchAgent, a new connector. Operating is knowing which lever to pull, when, without thinking. They feel similar because both happen at the keyboard, but they pull in opposite directions: building expands the surface area, operating demands you keep it small enough to hold in your head.

The failure mode is accretion. Every skill you install is cognitive surface area whether or not you use it — it shows up in the menu, it's one more thing to remember exists, it dilutes the half-second where you decide what to reach for. An unused skill is actually worse than no skill, because it costs attention while returning nothing. This is the software version of a junk drawer: the problem isn't any single item, it's that you can't find the scissors anymore.

Where it shows up in your workspace: /plan-day already pulls from learnings.md, four prefetched JSON files, two lesson renderers, and the Things connector — that's a lot of moving parts for one morning command, and it mostly works because each part does one thing. Contrast that with the skills list, which has grown by accretion with no prune pass. Operating the machine means periodically asking of each part: do I actually pull this lever, or is it just sitting in the menu?

Worked example

Inventory every skill available to you, with its one-line description, so you can see the whole menu at once instead of meeting skills one-at-a-time:

for f in ~/.claude/skills/*/SKILL.md; do
  name=$(basename "$(dirname "$f")")
  desc=$(grep -m1 '^description:' "$f" | cut -c14-80)
  echo "$name — $desc"
done | sort

# How many is that?
ls -d ~/.claude/skills/*/ | wc -l

The for-loop walks every installed skill folder and pulls the first line of its description — turning a scattered menu into one scannable list you can read top to bottom.
`cut -c14-80` trims the 'description: ' prefix and caps the length so the list stays one-line-per-skill and skimmable.
The final `wc -l` gives you the raw count — the number that usually surprises people. That number is your menu size, and menu size is the thing you're trying to shrink.

▶ Do it now

Run the inventory command above and read the whole list once, slowly — most of these you've never deliberately invoked.
Down the list, tag each skill in your head as Keep (pulled it in the last 2 weeks), Cut (never knowingly used it), or Hold (called by automation, not by you — like languages-render, learning-curriculum, plan-day).
Count the Cuts. Don't delete anything yet — just sit with the number. That count is how much menu clutter you've been carrying, and naming it is 80% of the win.

Gotchas

Don't `rm` a skill mid-sabbatical to 'clean up' — some are load-bearing for LaunchAgents (languages-render, learning-curriculum, and plan-day are all fired by cron, not by you). Cutting one silently breaks a 6:15am job. Archive to a _retired/ folder if you must, never hard-delete.
Bundled skills (gstack, superpowers, document-skills, plaud-*) arrive as plugins — you usually can't prune one card out of the deck, only disable the whole plugin. For those, the 'prune' is mental: know which three you actually use and ignore the rest.
Resist the urge to turn this audit into a build project. The goal is a shorter menu, not a new skill-manager tool — that would be the exact accretion trap the lesson is about.

Go deeper: Claude Code skills — official docs

One-card takeaway

A capability you never invoke isn't an asset — it's menu clutter that taxes every decision. The sabbatical win isn't the biggest toolbox; it's the smallest one that still does everything you actually do.

← Prev Index Next →