THE HONEYPOT / NO. 02

the honeypot / no. 02 / week of sep 28

effort is a dial, not a flex.

i save 20 to 50 new posts a week on ai, design and games. about 5 are worth stealing. here's the jar.

skip to the 5 picks ↓

the pile99 unfiled saves, newest first.

99
in the pile

99 saved, ~41 new this week, 5 kept.

5 worth stealing

drops #004 to #008

the best saves this week were about running agents with less. one shared brain, an effort setting, and a list of assumptions before any code.

#004@kloss_xyzarticle by @trq212

same task. 1 minute on low, 28 on max.

a pointer to the claude code effort article.

steal thisdefault to low or medium. go high once there's a plan.

list every assumption.

a one-paste assumption audit.

steal thismake the agent mark each assumption verified or guessed, then read the guessed column.

read it ↗used itthis page's build prompt opened with it

one brain before you switch models.

the shared-memory idea.

steal thiskeep one markdown notes vault that every agent reads, so context moves with you.

parallel subagents, not a bigger model.

a subagent usage recipe.

steal thisorchestrator on xhigh, one ticket and minimal context per subagent, a cheap model for the rest.

a brand system in, a motion video out.

claude design plus claude code making a branded video.

steal thisthe model can't hear, so have it measure tempo, loudness and drops before it cuts the sound.

what we built

assumption lists.

my build prompts now open with "list every assumption" before any code. the prompt that made this page did.

not sticky

a second agent-company os, faceless youtube autopilot, trading bots, logo-prompt spam, "full course" bait.

a new jar every monday.