Created and presented by Nino Chavez.

Session S12 · 7 chaptersScroll to read · use ← → to step between chaptersSkip to the record

WAYS OF WORKING · DEMO 12 NINO CHAVEZ

One component I didn't already have.

A coding-agent plugin with sixty-six thousand stars keeps arriving in my inbox. It advertises eleven agents and fifty-four hooks. Measured against the setup I already run, the version that could actually land on my machine came out at a delta of one.

"read through and research "oh my opencode" to understand its use case and how it compares to what i'm already doindg and if using it or adopting it is useful in my agentic ways of working." operator message, verbatim — note "compares to what i'm already doing," which is the whole job

or scroll to advance · to go back

02 THE REAL QUESTION

A feature list is half a subtraction

Eleven agents and fifty-four hooks is eleven and fifty-four against zero. Nobody who has been doing this a year is at zero. The number that decides adoption is what remains after your own setup is subtracted — and that half lives on your machine, not on the project's page.

Version one · runs on a different agent

The full plugin runs on OpenCode. I have it installed and have not opened it since September. My subscription is with a vendor whose terms that path violates — stated in the host agent's own documentation, not the plugin's.

Not a close call. Never the interesting half.

Version two · runs where I actually work

A smaller build installs into a terminal agent I drive daily. Eight components instead of the full surface. This one had a real path onto the machine, so this is the one worth measuring.

So the question stopped being "is this good" — by most accounts it is — and became "what does it add to this."

03 THE SUBTRACTION

Eight components, one column added

Project-rules injection

My agent already loads its instructions file. Mine is 253 lines

native

Work-continuation enforcement

I ported my own hook for this months ago; it is already running

native

An effort keyword

Duplicates native reasoning levels, mine enabled through the top tier

native

A Windows shell shim

I work on a Mac

n/a

A durable multi-goal loop

Disabled — its own tracker says the program it needs is not shipped

broken

Anonymous telemetry

On by default. A cost, not a feature

cost

A comment checker

New — and the piece users in the tracker name as a token waster

mixed

Language-server tools

Genuinely absent from my setup. The entire delta

new
3already running
1wrong platform
1broken upstream
2cost or mixed
1genuine gain

04 THE SOURCE THAT ISN'T CURATED

The documentation says what it does when it works

The issue tracker says what it does on other people's machines — 161 issues filed in thirty days, written by users, and impossible to quietly prune without the gap showing. It is the fastest read of a project's health that exists, and it is entirely mechanical.

"a simple 'Hello world' currently uses 15 to 25k tokens" — quoted back by the maintainer, who agreed: "It does appear to be using a lot because there are many tools currently provided." user report and maintainer confirmation, December 2025 — deferred loading was planned; I found no evidence it shipped

Scoping this honestly

That figure was measured on the full build with its whole tool surface. It does not transfer to the eight-component version, and I am not going to pretend it does. It tells you what the design optimizes for, not what the smaller build costs.

What does transfer

Context-consumption bugs were still being filed this week. The pattern is live even where the number isn't portable.

Read for shape, not for counts.

05 THE SHAPE OF THE FAILURES

It breaks when its host moves, and it doesn't fully leave

Dead after one host releaseplugin disappears
Works on one point version, broken on the nextagents stop loading
Task calls fail on a thirdtype error
Its own auto-update broke subscription authopen, filed two days ago
Uninstall leaves agents registeredopen · full build
Uninstall leaves the command and config behindopen · smaller build
Earlier in this session I called the smaller build "a defensible thing to just try," reasoning that trying was cheap and reversing was clean. The second half was wrong. Cheap-to-try is a claim about the exit, not the entrance. a recommendation retracted mid-session, by its own evidence

None of this makes it a bad tool. Its stability is downstream of a host that ships fast, which is a structural position rather than a defect. It does mean the cost of being wrong is an afternoon of cleanup, not one command.

06 THE FAILURE SEGMENT

My first answer was well-argued and under-researched

The first pass read the project's own pages — a polished README, reference documentation, a video the project links approvingly — and concluded that independent evidence was scarce. It wrote that conclusion up as the finding.

What was wrong with it

Not the reasoning. The research. Every source it examined was one the project published or selected, and it treated the resulting thinness as a discovery rather than as a symptom of where it had looked.

What the operator said

"this is not something we should ship. we should solve for getting the needed context." The artifact came down. The second pass ran three queries against the tracker and had better evidence in about a minute.

A gap is a task, not a thesis.

The tell is worth keeping: when a research pass concludes that the evidence does not exist, that is a claim about the search, and it should be checked before it becomes a claim about the world.

07 YOUR VERSION OF THIS

Run the subtraction, not the feature list

Write your baseline down

You cannot measure an addition without the starting number. Mine is a public repository, which is the only reason this took twenty minutes rather than an afternoon. Do it once and every future tool question becomes a lookup.

Sort into three buckets

Native, new, and blocked — new but unusable for reasons unrelated to quality. People collapse the third into the other two, which distorts the answer both ways. An excellent component you can't run is not a gain and not a criticism.

Query the tracker, then the exit

Volume, filing rate, and the failures specific to your platform and your host's version. Then search the tracker for the uninstall itself. Open reports that removal leaves residue are worth more than the whole install guide.

a feature list is half a subtraction native · new · blocked the tracker is the uncurated half of the repo check the exit before the entrance
Colophon. One session, July 2026: one operator prompt with four links in, a published artifact taken back down, and a second pass that changed the answer. Component mapping was run against a real machine's configuration; issue counts, quotes, and dates come from the project's public tracker and are dated where they are stale. The token figure is scoped to the build it was measured on. The forum thread the first pass chased is still unread — it stopped mattering once 161 monthly filings turned up. Verification as a discipline is demo 07; provenance labels for research output are the applied card; auditing your own setup rather than someone else's is demo 11. Companion method card: Run the Subtraction Before You Install.

What happened

One Component I Didn't Already Have

A plugin advertising eleven agents and fifty-four hooks, measured against the setup I already run. Of the eight components that could land on my machine, three were already going, one was the wrong platform, one was broken upstream, two were costs. The delta was one.

Collection
Ways of Working
Record
S12
Format
Session
01 / Reader

Who this is for

Anyone handed a feature list and asked whether to adopt it — especially with a year of accumulated hooks, rules, and helpers already loaded.

02 / Evidence

What the session shows

A component-by-component subtraction against a real machine, three mechanical queries against the part of a repo nobody curates, a retracted recommendation, and the correction that produced all of it.

03 / Practice

What to reuse

Write your baseline down, sort the tool's list into native, new, and blocked, then search its tracker for the uninstall before you run the install.

Applied techniqueRun the Subtraction Before You InstallA feature list is half a subtraction. Enumerate your baseline, sort the tool's list into three buckets, then search its tracker for the uninstall.