The most useful thing my agent does is refuse
Refusals are only useful if they name the remedy. Four that fired during one session, what each redirected me to, and the one that fired on nothing at all.
All posts
The agent that writes the check that blocks it
When the same agent writes the fix and the gate, the gate agrees by construction. RED before GREEN is what makes a machine-written check mean anything.
An agent that installs the app and looks at the screen
The capture loop that makes an on-device claim checkable: force-stop, install, launch, capture, hash. Why the order is load-bearing and what it cannot see.
The context window is not your constraint — what you put in it is
Scoping a shared instruction bundle saved 8.2% per worker, against 40% guessed. Why residency beats capacity, and the surface that was never loaded at all.
Put the control flow in code, not in the prompt
Sequencing belongs in a script and judgement in the model. One routine commit cost 180k tokens as reasoning and seconds as bash. Here is the line.
A hinge is a physical object, and your layout should know where it is
Foldables give two things a resize doesn't: a posture and a hinge with real coordinates. Handling them well means avoiding the seam and using tabletop mode, not just supporting a new width.
Where to put the human
Gate placement for autonomous agents. Reversibility is the criterion, not confidence, and the consent surface matters as much as the question.
MCP, or a shell script? The honest comparison
Decided on concrete cases rather than preference. Where a protocol server earns its keep, where a script wins, and the failure that shows up unattended.
I swept the parameter instead of arguing about it
A threshold argument settled by measuring four values over the real corpus. The knee, the trade that stops paying, and why the numbers go in the comment.
The same rule in two files is how agent-written code drifts
A protected-path list was typed by hand and had drifted nine entries behind reality. Derive the list instead of copying it, and gate the copies you keep.
One level of fan-out, and why that is enough
A subagent cannot delegate, so the top session is the only scheduler. Designing around that, and why a barrier between stages costs more than it looks.
The prompt is not the artifact — the runtime is
A behaviour defect belongs in the command's runtime file, not in better instructions next time. What to edit, how to lock it, and where that stops working.