ajitta

engagement-retention-advisor

Produce integrated engagement proposals connecting an in-session moment to a next-visit reason (재미도 살리고 다시 오게) — guild/clan, UGC and creation-sharing, meta-progression as a system, in-game economy — for games, interactive narrative, fortune/saju, AI companion, journaling, learning and similar consumer interactive apps. Use when the request carries TWO deliverables — moment design AND lifecycle/return strategy — or the moment-to-return link itself is the question ('what in-session moment raises retention?'), or a moment complaint is paired with a churn complaint ('combat feels flat and players churn'), or the ask is to compare, sequence or prioritize a moment-level against a lifecycle-level investment, or an engagement complaint names no scene and no metric ('our players are bored', 'make it more addictive', '재미없대요', 'the game feels grindy'), or the goal itself is raw time-in-app or session length — start here and ask one scoping question. A retention metric or a return cited only as motivation is NOT a second

1 Updated 2 days ago

interaction-reward-moments

Design and improve in-session reward moments — the reveal, choice, feedback, anticipation and game-feel beats inside one session (손맛, 타격감, 연출, 밋밋하다, 도파민 포인트) — for games, interactive narrative, fortune/saju, AI companion, journaling, learning and similar consumer interactive apps. Use when the deliverable is a named scene, loop, screen or beat: 'the level-complete screen feels hollow', 'make the gacha pull land', 'the tutorial's first win is flat', or a short-session / quit-mid-session complaint (session length is a diagnostic, never a target). Also owns the strength, safety and accessibility of a feel effect — flash, shake, haptics, camera motion — as game feel, not general UI or visual design. A retention metric or a return cited only as motivation ('D7 is low, fix the reward staging') does NOT disqualify this skill. Route away when the deliverable is the cadence or schedule of a battle pass, login rewards, streak or daily quest, to retention-strategy-designer — but the staging of any one of those beats (th

1 Updated 2 days ago

retention-strategy-designer

Diagnose and improve lifecycle retention (리텐션, 이탈률, 잔존율, 복귀 유저) for games and consumer interactive apps: cohorts, D1/D7/D30, activation, onboarding/FTUE funnel (which step players drop at, gating, D0→D1 leak), habit, churn, win-back of lapsed or dormant players, and every multi-day cadence system — battle pass, season, login rewards, streaks, daily quests, LiveOps calendar, meta-progression pacing and unlock schedule. Also owns retention metric definitions (classic vs rolling day-N), reading a pasted cohort table or curve, instrumentation and event taxonomy, A/B and experiment design, notification copy, and — only when retention is the constraint being protected — the monetization-versus-retention trade-off (ARPDAU/LTV vs D7, ad load, offer cadence, paywall). Do NOT trigger when the deliverable is one in-session scene, reveal or staging fix, or a session-length complaint — even if D1/D7/onboarding is cited as the motivating metric; use interaction-reward-moments. Do NOT trigger when moment design is a second

1 Updated 2 days ago

blindspot

Blind-spot pass — investigate unknown unknowns before implementation and turn them into a better prompt. Use when the user says "blind spot pass", "what am I missing", "unknown unknowns", or before work in an unfamiliar codebase area. Not for routine, well-understood tasks.

brainstorm

Brainstorm the intervention — spread ~10 codebase-grounded candidates from ship-this-afternoon to quarter-long bet. Use on "brainstorm interventions", "show me options", or a problem named with no approach chosen. NOT for UI/design variants — use prototypes.

buy-in

The buy-in doc — a ship-readiness pitch that leads with a working demo, pre-answers reviewer objections with evidence, states limitations, and names who must sign off. Use when the user says "buy-in doc", "prep me for review", or before requesting review/approval/merge.

interview

Pre-implementation interview — the model asks the user the questions that turn known unknowns into decisions, biggest blast radius first. Use when the user says "interview me", "ask me questions before implementing", or requirements are incomplete and implementation has not started.

loop

Run the know-your-unknowns workflow end to end (explore → question → plan → implement → verify → quiz) on significant or ambiguous work. Use on "unknowns loop", "run the operating loop", "know your unknowns". A bare "loop" means the built-in interval runner, not this skill.

notes

Implementation notes — log significant deviations from the plan while working. Use automatically when implementation hits an unknown the plan did not cover, and on "implementation notes", "record a deviation", "where did we diverge from the plan?".

plan

The tweakable plan — a plan sorted by likelihood-of-tweaking, not execution order, so the most changeable decisions get reviewed first. Use on "plan this", "make a plan", "tweakable plan". Not native plan mode: writes a reviewable pre-implementation document.

prototypes

Divergent prototype fan-out — N (default 4) prototypes with clearly different design philosophies to react to. Use when the user says "divergent prototypes", "design options", or cannot say what they want ("know it when I see it") for a UI, dashboard, or API design.

quiz

Post-work comprehension quiz — explain what changed, then quiz the user so they can represent the work in a PR or handoff. Use when the user says "quiz me", "test my understanding", or right before creating or merging a PR for work the model largely implemented.

reference

Point at a reference — treat a provided example as a map, not an answer, and prove comprehension before implementing. Use when the user says "use this as a reference", "make it like this", or supplies example code, a mockup, or a competitor's UX to build from.

teach-me

Teach me my unknowns — an interactive explainer for a domain's vocabulary and mental model, so vague requests become precise. Use on "teach me", "make me an explainer", or when the user cannot name what they want. Blindspot investigates a codebase; teach-me teaches vocabulary.