WAYS OF WORKING · APPLIED NINO CHAVEZ
Run the subtraction
before you install.
Four checks that run before anything touches your machine: enumerate what you already have, subtract it from what the tool ships, query the part of the repository nobody curates, and confirm the uninstall works.
→ or scroll to advance · ← to go back
02 THE FAILURE THIS FIXES
A feature list is half a subtraction
Fifty-four hooks is fifty-four against zero. Against a setup that has been accumulating for a year, the honest number is whatever remains after your own is subtracted — and nobody publishes that half for you, because it lives on your machine.
What it feels like
An impressive list of capabilities, each one plausible, adding up to an obvious yes. The volume does the work that measurement should be doing.
What it is
An unfinished equation. The same list produces a different answer for a person running stock tooling than for a person with a year of hooks and helpers already loaded.
Cost of skipping: you adopt on volume.
03 CHECKS ONE AND TWO · THE SUBTRACTION
Write your baseline down, then sort into three
Check one is the one people skip and the one that pays permanently: get everything your agent loads at startup into a list you can read. Do it once and every future tool question becomes a lookup.
Native
Your setup or your agent already does this — remove it from the gain column entirely
New
Genuinely absent. This is the delta, and the only column that argues for installing
Blocked
New but unusable for reasons unrelated to quality — wrong platform, broken upstream, needs a subscription you don't hold
Blocked is the bucket people collapse, and it distorts the answer both ways. A component that is excellent and unusable is not a gain, and it is also not a criticism of the tool. Then run it backwards: what does your baseline do that the tool's list never mentions?
curl -fsSL https://raw.githubusercontent.com/<owner>/<repo>/<branch>/docs/reference/features.md ls ~/.claude/skills/ ~/.claude/commands/ ~/.claude/hooks/ # your side
04 CHECK THREE · THE UNCURATED HALF
Documentation says what it does when it works
The issue tracker says what it does on other people's machines. Users write it, and unlike a testimonial section it cannot be quietly pruned without the gap showing. Three queries: volume, arrival rate, and your own failure modes.
gh api "search/issues?q=repo:<o>/<r>+is:issue+is:open&per_page=1" --jq '.total_count' gh api "search/issues?q=repo:<o>/<r>+is:issue+created:>=<30-days-ago>&per_page=1" --jq '.total_count'
gh api "search/issues?q=repo:<o>/<r>+is:issue&sort=comments&order=desc&per_page=25" \ --jq '.items[] | "\(.comments)c \(.state) #\(.number) \(.title)"'
Read for shape, not counts — a high open count on a popular project mostly means it is popular. Does it break when its host releases? Search the host's version numbers. Do its own updates break things? Search after updating. Are the busiest threads features, or the same bug returning?
05 CHECK FOUR · THE EXIT
"Cheap to try" is a claim about uninstalling
Every installer is one command. The exit is the part nobody tests, and it decides whether being wrong costs you a minute or an afternoon.
Read what it writes
Install guides usually say what lands on disk. Check each location against your machine before running anything. Prefixed names and new named sections are additive and reversible; bare generic names and rewrites of sections you already use are not.
Then search for the uninstall
Almost nobody does this. Open reports that removal leaves things behind are worth more than the entire installation guide, because they tell you the true cost of being wrong.
No answer is itself the finding.
gh api "search/issues?q=repo:<o>/<r>+is:issue+uninstall+OR+cleanup+OR+remove" \ --jq '.items[] | "\(.state) \(.created_at[0:10]) #\(.number) \(.title)"'
06 START HERE
One tool, four checks, forty minutes
Minutes 0–10
List what your agent loads at startup, and name what each item does in a sentence without looking it up. Anything you cannot explain is a finding — it loads every session and you do not know why.
Minutes 10–22
Pull the tool's reference enumeration and sort every component into native, new, or blocked. Then run it backwards. Count the new bucket — that count is the decision.
Minutes 22–40
Three tracker queries plus two searches specific to your platform and host version. Then the uninstall search. Write one sentence on the recurring failure mode and whether it would reach you.