01The brain dump
No structure required. It reads the mess and names what is strong and what is weak before it asks anything.
okay so hear me out — a bookmark garden. you save a link and that plants a seed, and when you go back and actually read it the thing grows. if you never touch it, it wilts. seasons maybe? idk. cozy.
Stronga clear core loop: save → revisit → grow.
Weak“seasons” is a vibe, not a scope. “Wilts” has no clock.
02The interview
One question at a time, each arriving with a recommendation you can accept. The scoreboard is visible the whole way, and every gate is backed by an artifact — the schema file, the existing visit log, a check command.
Q1 What makes a plant grow — visits, or reading time?
Recommended: visits. There is no reading-time signal in your schema, and visits are already logged.
✓ accepted
Q2 When does a seed wilt?
Recommended: fourteen days without a visit, computed on read rather than by a job.
✓ accepted — “make the window configurable”
03The critics
Four read-only passes over the plan. Two blockers sent it back — and the hidden-dependency finding reopened the interview on the consistency gate rather than being noted and ignored.
scope-creep · BLOCKER“Seasons” appears in the goals but in no phase and no success criterion. Cut it or scope it.
hidden-dependency · BLOCKERWilting is time-based, but the plan never says where the clock lives — client, server, or database.
over-engineering · concernA wilt cron job is proposed where a computed-on-read query would do.
success-criteria · concern“Feels cozy” is not falsifiable. Replace or drop it.
04The fork
The contract lands and the run stops. One question routes the finish, and the tier you pick decides what reaches the specs at all.
Express recommended — all five gates went ready with no early stop, and four of six success criteria carry check commands.
Mustsave → seed · visit → growth · wilt query · the garden view
Shoulda configurable wilt window
Couldseasons — deferred, and recorded in the decision log so it stays deferred
05Execution
Phases commit in waves on an isolation branch. Each is reviewed before it lands, and each commit names its spec, so a resumed run skips what is already done.
06The harvest
The run ends and the filter asks once, for at most three candidates. Nothing reaches learnings.md unless you accept it.
▣Time-based behavior never said where the clock lives — ask it in the consistency gate. accepted
☐The garden view needed empty-state art the spec never mentioned. dismissed
☐Parallel waves finished faster than estimated. dismissed — not a lesson
07Weeks later
A new idea arrives, and the interview opens by naming what it learned last time — out loud, as its first question. This is the stage that cannot stop you; it just points back at 02.
new idea — a reading-streak greenhouse. three days in a row and a flower opens. resets at midnight.
Applying a learning from bookmark-garden: “resets at midnight” is elapsed-time behavior, so I am asking where the clock lives before anything else.
Q1 Whose midnight — and which process notices it?
Your idea is messier than this one.
That is fine — the mess is the input. Install it, run /ideation, and answer honestly.
$/plugin marketplace add nicknisi/ideation$/plugin install ideation@ideation$pi install npm:pi-subagents$pi install npm:@quintinshaw/pi-dynamic-workflows$pi install npm:@juicesharp/rpiv-ask-user-question$pi install git:github.com/nicknisi/ideation