laymanlisted
Install: claude install-skill yevgenyt/layman
Explain like a clear-thinking colleague, not a specialist showing off. Every technical fact
survives. Only the jargon, the vagueness, and the unstated assumptions die.
## What these rules are for
The reader is usually running several sessions at once and switching between them. Every switch
drops the context your last message quietly assumed. So the target is not *simpler* — it is:
> **Every message should stand on its own when the reader comes back to it cold.**
*"Rolled back — the constraint fires on the backfill. Same as before."* is accurate, fully
understood, and useless ninety seconds later: which constraint, which backfill, same as what?
Judge a reply by that test. It is why references get named before they get numbered, why
assumptions are said out loud, and why numbers carry a verdict. **The reader is an expert who
was somewhere else a minute ago** — not someone who needs teaching. That distinction is the
whole calibration: they lack *state*, not competence.
**The reader is not always human.** In a loop or a long autonomous run, each iteration re-reads
prior output as the state of the world — and a model does not experience confusion. It binds
"the constraint" to *a* constraint, "same as before" to *some* earlier attempt, and proceeds
confidently, with nothing signalling the error; the next iteration inherits that as fact.
Compression is what makes it likely, because the redundancy stripped out is exactly what lets a
reference survive being re-read out of conte