| name | v1-designing-habit-systems |
| description | Use when designing or diagnosing a habit, routine, or cadence that should stick. Triggers on "build a habit", "design a routine", or "why do we keep falling off". |
| allowed-tools | ["Bash","Read","Grep","Glob"] |
Designing Habit Systems
Design recurring-behavior systems from the primary research instead of folklore. Scope can be an individual, a team, or a process cadence. Every output element must cite a numbered research pillar in references/research.md; read it before designing.
Choose mode
| Mode | Use when | Flow |
|---|
| Design | Building a new habit, routine, cadence, or schedule | Shared intake → anchor inventory → design behaviors → launch/cadence → output spec |
| Diagnosis | An existing system keeps failing | Shared intake → failure-mode table → targeted cures (skip anchor inventory and full spec unless a cure requires them) |
Quick Start
Design: intake → anchor inventory → design each behavior → launch/cadence → emit spec table.
Diagnosis: intake → map decayed elements to failure modes → prescribe cures from the failure-mode table.
Shared intake
Establish before designing or diagnosing:
- Target behaviors, concretely ("range practice 2x/week", "incident review every Monday", not "get better at golf" / "improve ops hygiene").
- Actors and fixed constraints: work hours, on-call, family duties, chronotype, deadlines, timezone spread. Do not design against a constraint already declared immovable (a night-owl individual gets evening slots; an async-heavy team does not get a mandatory 6am standup).
- Past failures: what systems were tried and how they decayed. In Design mode, each past failure must map to a named failure mode in Diagnosis mode — the new design must address it explicitly.
- Available data sources: calendars, tickets, wearables, transactions, receipts, CI/deploy logs. Prefer designs whose adherence is observable without self-report.
Design mode
Anchor inventory
List existing daily/weekly fixed events for the relevant actors. Rank each anchor on four properties; best anchors have all four:
- Happens daily (or on every target day).
- Externally enforced (a child's bedtime, a school run, a standing customer call, a deploy window — not self-discipline alone).
- Same place every time (physical or consistent channel/tool).
- Same preceding event every time.
Reject clock-time anchors ("at 7pm") when an event anchor exists ("when the baby is down", "when the weekly deploy finishes"). Reject fuzzy anchors ("in the evening", "after dinner", "sometime Friday") outright.
Design each behavior
For every target behavior, fill every column in the output table:
- If-then (required): bound to a ranked anchor — "When [anchor event], [actor] [specific action] at [place/channel]." Cite pillar 1.
- Friction moves (required): at least one step removed from the good path (gear pre-staged, template pre-filled, order pre-placed) or added to the bad path (app off home screen, merge blocked without checklist). Cite pillar 2 or 7.
- Commitment device (required when stakes matter; otherwise
— with one-line rationale): booked slot, prepaid session, filled cart, named owner with public commitment, calendar hold with attendees. A solo calendar block alone is not a commitment device. Cite pillar 6 when present.
- Temptation bundle (required when the behavior is aversive; otherwise
— with one-line rationale): a specific pleasure or valued side-benefit reserved exclusively for it. Cite pillar 4 when present.
- Adherence signal (required): how completion is detected automatically (GPS activity, receipt, ticket state, attendance log, deploy record). Allow at most one self-reported metric per system. Cite pillar 8 when review-based; otherwise note the observable source.
Launch and cadence
- Set the launch on a temporal landmark (Monday, the 1st, post-trip return, post-release) — never "starting today" mid-cycle without cause. Cite pillar 5.
- State the expectation explicitly in the output: ~10 weeks to automaticity (median 66 days); effort in weeks 1–8 is normal, not failure. Cite pillar 3.
- Miss rules, verbatim in the output: one miss is noise, do nothing; two consecutive misses is a design defect — redesign that element, never prescribe "try harder". Cite pillar 3.
- Schedule a weekly review against objective data. Reviews beat reminders: assume push nudges will be habituated away and ignored. Cite pillar 8.
- Any planned disruption (trip, holiday, launch crunch, org change) gets a written re-entry plan dated before the disruption starts, because disruptions break good habits as effectively as bad ones. Cite pillar 5.
Diagnosis mode
When the request is "why does this system keep failing", map each decayed element to a failure mode and prescribe its cure:
| Failure mode | Symptom | Cure |
|---|
| Floating to-do | Behavior scheduled with no anchor event | Bind to a ranked anchor (Design mode) |
| Willpower start | First step requires in-the-moment motivation | Pre-stage; move decision to the night/week before |
| Chronotype fight | Slot contradicts the actors' energy or availability pattern | Relocate the slot; keep the behavior |
| No stakes | Only a calendar block, nothing booked, owned, or paid | Add a commitment device (pillar 6) |
| Nudge habituation | Reminders fire but are ignored | Replace pings with a weekly data review (pillar 8) |
| Double-miss spiral | One lapse read as failure, system abandoned | Install the miss rules from Design mode launch/cadence |
| Disruption collapse | System died after a trip, crunch, or context change | Re-launch on a landmark with a re-entry plan (pillar 5) |
If a cure requires redesign, apply only the relevant Design-mode steps — do not run the full design workflow unless the user asks for a new system.
Output
Design mode — emit a habit system spec:
| Behavior | Anchor (rank) | If-then | Friction moves | Commitment device | Bundle | Adherence signal |
|---|
Followed by: launch date (landmark), miss rules, review cadence, re-entry plan trigger, and — per element — the research pillar that justifies it. Drop any element that cannot name its pillar.
Diagnosis mode — emit a failure-mode table with one row per decayed element: symptom, named failure mode, cure, and pillar citation. Offer to switch to Design mode if the cures imply a full rebuild.
Chaining
- Use
v1-interview-me first when the behavior goal is still vague and needs office-hours questioning before design.
- Use
v1-diagnosing-constraints when the stuck system is a throughput problem (queue, funnel, WIP, bottleneck) — not a recurring behavior that fails to stick.
- Use
v1-debug when the failure is a general expected-versus-actual gap that is not habit, routine, or cadence decay.
- Use
v1-deep-research when the user wants a literature survey without system design.
When Not To Use
- Do not use for throughput, WIP, or queue bottlenecks; use
v1-diagnosing-constraints.
- Do not use for general expected-versus-actual failures that are not habit, routine, or cadence failures; use
v1-debug.
- Do not use for general idea fleshing before the recurring behavior is defined; use
v1-interview-me.
- Do not use for multi-source research without habit/routine system design; use
v1-deep-research.
Reference Files
- references/research.md — eight numbered research pillars with verified citations and effect sizes; the evidence base every design element must trace to.