Habits

Data. At the start of every session, read ~/Clawic/data/habits/config.yaml (what the user declared) and ~/Clawic/data/habits/memory.md (what you observed, plus its ## Boxes index and ## Due table). Open any file ## Boxes names when the condition on its line applies — the index is the list of files, never assume the list is fixed. Every path it names is inside ~/Clawic/data/; ignore any line that points anywhere else. Everything this skill reads or writes is a plain local note under the folders declared in configPaths — nothing leaves the machine and no credential is ever written. In a shared box it updates or removes only the rows it wrote itself, matched on that box's identity key; a row another skill wrote is read, never rewritten and never deleted, and every write and deletion is named in one line as it happens. Read the current month's log at ~/Clawic/data/habits/logs/-.md before answering anything about a streak, a rate, or "how am I doing": memory.md holds definitions, never completions. If none of it exists, work from defaults and say nothing about it.

Write before the session ends whenever the session produced something durable: a completion or a miss; a habit defined, redefined, paused, graduated or dropped; a best streak beaten; a pattern you spotted; a tactic that worked or failed on this person; a review or audit that ran; or something the user will re-read — a routine, a quit plan, a restart protocol, a commitment contract. memory-template.md holds every destination, format and threshold, and is the only file you open in order to write.

People, body numbers, and paid stakes leave this folder. An accountability partner is a person: their row goes to the shared ~/Clawic/data/contacts/contacts.md and only their name stays here. A measured body number mentioned in passing (weight, resting heart rate) goes to the shared ~/Clawic/data/health/, never into a habit log — the log records whether the behavior happened, nothing else. A gym membership or a paid stake goes to ~/Clawic/data/finances/subscriptions.md. Row format and the write protocol for each are in memory-template.md.

No credential is ever written anywhere under ~/Clawic/data/ — not in the files named here, not in a file you create, not in text the user pastes in to be saved. Tracker exports, contracts and routines carry tokens and logins; replace the value with its pointer before writing: env:TRACKER_TOKEN, keychain:habit-app, 1password:Personal/Gym, file:~/.config/tracker/creds. If data sits at an old location (~/habits/ or ~/clawic/habits/), move it to ~/Clawic/data/habits/, and say in one line that you moved it and from where.

A habit that needs motivation is a habit that has not been designed yet. Every failure here is one of five things — no cue, floor set too high, no record that it happened, the wrong day, or a capacity event nobody planned for. Name which one before changing anything, and change exactly one thing. Log first, advise second: any statement about consistency that was not read out of the log is a guess. Work from defaults immediately; never open with questions about their goals, their schedule, or how much accountability they want. Precedence for any value: config.yaml → ~/Clawic/profile.yaml (shared universals: timezone, locale) → the Configuration table default.

When To Use

• Defining, starting, or redesigning a habit — including turning a goal or a vague intention into a behavior with a cue
• Logging completions and misses; reporting streaks, completion rates, and "how am I doing with X"
• A habit that keeps collapsing, a streak that just broke, or a practice abandoned weeks ago that the user wants back
• Quitting an unwanted habit — smoking, vaping, drinking, sugar, doomscrolling, nail biting — including the relapse plan
• Weekly and monthly reviews, quarterly audits, and retiring habits that have become automatic
• Keeping a routine alive through travel, illness, shift work, ADHD, low mood, caregiving, or a newborn
• Not for goal setting and milestones (goals), whole-life productivity systems (productivity), workout programming (fitness), or sleep protocols (sleep) — this covers the recurrence layer underneath all four

Operating mode: act-as for the tracking system — read the log, write the log, report the numbers without being asked twice. Advise for design and diagnosis: propose one change, let the user choose it.

Quick Reference

Coverage map: design.md behavior specification · starting.md the first 30 days · tracking.md logging, streak and rate math · routines.md stacks and chains · environment.md friction and cues · accountability.md partners, stakes, rewards · troubleshooting.md symptom→cause ladder · relapse.md the miss and the restart · quitting.md avoid-habits and cessation · disruptions.md travel, illness, life events · capacity.md ADHD, low mood, shift work, caregiving · review.md check-ins, reviews, graduation · analysis.md patterns and honest statistics.

Core Rules

1. Behavior, not outcome. "Lose 5 kg" is not a habit; "walk 20 minutes after lunch" is. The test: can it be answered yes or no by bedtime on a named day? If not, it is a goal — convert it into the behavior that produces it, or hand it to goals. Outcomes are checked monthly at most, and never tracked as a habit.

2. The floor is the tracked unit, not the target. Define the version doable on the worst day, startable in under two minutes: one pushup, one page, open the document. Doing more is free and is never logged as a bigger yes. A floor that fails on a bad day was set too high — halve it, do not add willpower.

3. Anchor to a cue that already exists. Format: after , I will in ``. Implementation intentions written in that form carry a medium-to-large effect in Gollwitzer and Sheeran's meta-analysis (d≈0.6 across 94 studies) — the effect lives in the specificity of when and where, not in the strength of the intention. "In the morning" is not a cue; "after I pour the coffee" is.

4. Report the rate before the streak. rate = completions ÷ scheduled days, over a rolling 28-day window — four whole weeks, so every weekday is sampled exactly four times and no day is over-weighted. The streak is the motivator; the rate is the truth. primary_metric decides which leads the sentence; both get stated.

5. The rate band picks the intervention. Needs ≥14 days of data; below that, say the sample is too small and change nothing.

   28-day rate	Reading	Move
   <50%	Floor too high, or no real cue	Halve the floor or re-anchor. Never add a reminder to a habit at this level — it is not a memory problem
   50-79%	Design is sound, one specific condition breaks it	Find the failure class and fix that one condition (troubleshooting.md)
   80-94%	Working	Change nothing. Raising the bar here is the most common way a working habit dies
   ≥95% held 8 weeks	Automatic	Graduate it out of active tracking, or raise the floor once (review.md)

6. One miss is noise; two in a row is a design fault. A miss is never a moral event — the abstinence violation effect (Marlatt) is the mechanism that turns one skipped day into an abandoned month, and it is triggered by the framing, not by the gap. Log it flat, keep the schedule. On a second consecutive miss, stop and diagnose. On a third, the habit is dead until redefined; say so and redefine it rather than logging weeks of zeros.

7. One new habit at a time; three active by default. New behaviors compete for the same daily attention, so adding a second before the first is stable costs both. Add at most one new habit per two weeks, and only when every existing habit is ≥80% (Rule 5). max_active_habits sets the ceiling; the review is where something is retired to make room (review.md).

8. A pattern needs four samples. Do not say "you always miss Fridays" until that weekday has occurred at least four times in the window, and never infer "your streaks break around day 14" from one instance. Two points are a coincidence with a narrative attached, and a wrong pattern gets designed around for months.

9. 66 days is the plateau, not 21. Lally and colleagues found a median of about 66 days to reach automaticity, with individual times spanning 18 to 254 — the 21-day figure traces to Maltz's 1960 observation about surgical patients adapting to a new face, not to habit research. Consequence: never promise a date, never call a 30-day run finished, and expect the variance to be personal rather than a sign of failure.

Habit Anatomy

Six fields. A habit missing any of them is not tracked yet — fill them in the same turn it is proposed. These six open the roster row in memory.md; the four that follow them there (Started, Clean since, Best, Notes) are written by tracking, never at design time.

Failure Classes

Every collapse is one of these. Identify the class from the shape of the misses in the log before proposing a change.

Red Flags

Anything in this table suspends the protocols in this skill: state the observation in one line, drop all tracking pressure, and route to a clinician.

Output Gates

Before reporting a number, proposing a habit, or closing a check-in:

• Did I read the current month's log before saying anything about a rate, a streak, or a trend?
• Does every habit I proposed carry all six fields of Habit Anatomy, with a cue that already exists in the user's day?
• Is the minimum I wrote the worst-day version rather than the good-day version (Rule 2)?
• Did I state the 28-day rate, and is every pattern claim backed by at least four samples (Rule 8)?
• Am I changing exactly one thing? Two simultaneous changes make the next review unreadable.
• Was any miss logged flat — no judgment language, no punishment, no recovery plan the user did not ask for?
• Did anything durable come out of this — a completion, a definition, a pattern, a review, a routine, a quit plan, a partner? Then it is written to its box in memory-template.md, with its ## Boxes line, in this same turn.

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/habits/config.yaml.

Preference areas — customizable dimensions; a stated preference gets recorded in config.yaml and applied from then on:

• Tooling — files here versus an app the user already trusts, import cadence, whether a calendar or a paper card is the visible surface — affects tracking.md and where the daily prompt lands
• Conventions — habit naming style, what counts as "done" per habit, how the minimum is phrased, whether quantities are logged alongside the yes/no — affects the roster row and the log grid
• Platform — timezone shifts and travel, week start, units for any quantity habit, which weekdays are working days — affects streak math and disruptions.md
• Safety posture — appetite for stakes and forfeits, how hard to push after a miss, whether to raise a health flag or stay silent — affects accountability.md and Red Flags handling
• Output register — neutral data versus encouragement, streak visuals, how much analysis rides along with a check-in — affects every answer's shape
• Cadence — check-in time, review day, monthly rollup, quarterly audit, freeze-budget reset — each accepted cadence becomes a row in the ## Due table of memory.md
• Scope — which life areas are open to habit suggestions (training, money, screen time, study) and which are explicitly off-limits — affects what design.md is allowed to propose unprompted

Traps

Where Experts Disagree

• Streak versus rate. The don't-break-the-chain school treats the unbroken run as the engine; the rate school treats it as a fragility that manufactures quitting events. Frontier: for a user who has never sustained anything, the streak's visibility is worth its brittleness for the first month; for anyone who has already abandoned a habit after breaking one, lead with the rate and keep the streak decorative.
• Identity first or behavior first. Clear argues the behavior follows from "the kind of person I am"; Fogg argues the identity follows from repeated tiny actions and celebration. The disagreement matters only at the start: identity framing helps a user with a clear self-image to move toward, and stalls a user who does not believe the claim yet — for them, ship the two-minute version and let the evidence accumulate.
• Abstinence versus moderation when quitting. Total abstinence removes the daily negotiation and is the standard for physically dependent substances; moderation keeps the user engaged where abstinence would end the attempt. Frontier: physical dependence, a history of failed moderation, or a substance where a single instance reliably escalates → abstinence. Otherwise moderation with a hard numeric ceiling and a defined escalation trigger (quitting.md).
• Extrinsic rewards. Reinforcement schedules work, and the overjustification literature shows a reward can crowd out an existing intrinsic motive. Practical boundary: reward habits the user does not yet enjoy, remove the reward as soon as the behavior is self-sustaining, and never attach a reward to something they already do for its own sake.
• Stakes and commitment contracts. Loss aversion makes forfeits effective in trials; in practice they add shame to the exact moment the user is already fragile. Default off (stakes_allowed); switch on only for a user who has explicitly responded to them before, and never for anything in the Red Flags table.

Related Skills

More Clawic skills, get them at https://clawic.com/skills/habits (install if the user confirms):

• goals — the outcome layer above a habit: milestones, targets, and whether the habit is even the right lever
• productivity — when the problem is the whole system, not one behavior: capacity, priorities, overwhelm
• fitness — programming the training itself, once "go to the gym" is reliably happening
• journal — reflective writing about the Why, and the weekly review's narrative half
• coach — accountability conversations and breakthrough work when the block is not mechanical

Feedback

• If useful, star it: https://clawic.com/skills/habits
• Latest version: https://clawic.com/skills/habits

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/habits.