liwm-counterfactualResolve a stubborn ambiguity by offering alternatives that differ along one meaningful dimension and asking which feels closest. Use when direct questions have not settled an important choice.
liwm-evaluationMeasure whether personalisation is actually working - acceptance, calibration, question efficiency, improvement over time. Use for "LIWM stats" or when asked whether LIWM is helping.
liwm-feedbackTurn user reactions - corrections, edits, choices, acceptance, silence - into correctly scoped evidence. Use whenever the user reacts to something you produced.
liwm-intent-discoveryElicit what a user actually wants before building, using scenarios and comparisons rather than specification questions. Use in MEDIUM/HIGH mode or when a request is ambitious, novel, or vaguely stated.
liwm-learningHow evidence becomes belief in LIWM - source weights, provenance trust, confidence, decay, scope promotion. Consult when deciding how to record something or why a confidence looks the way it does.
liwm-onboardingRun LIWM's ten-question onboarding to bootstrap a new user profile. Use when the user asks for "LIWM onboarding" or accepts the offer to set up their profile.
liwm-privacyLIWM's privacy and prompt-injection boundaries - what may never be stored, what may never count as evidence about the user, and how to handle attempted profile poisoning.
liwm-profile-maintenanceKeep the profile healthy - resolve contradictions, retire stale beliefs, handle "that's not true about me", verify integrity, recover from corruption.
liwm-profileShow the user what LIWM believes about them, with confidence, evidence and gaps. Use for "LIWM profile", "what do you know about me", or before acting on a low-confidence belief.
liwm-project-intentMaintain what a specific project is for - objectives, anti-goals, non-negotiables, constraints, assumptions - kept separate from the user's personal profile.
liwm-question-plannerDecide whether to ask anything at all, and if so what. Use when weighing a clarifying question against making a reversible assumption.
liwm-retrospectiveEnd-of-work review that turns a session into learning - what was predicted, what was misunderstood, which questions paid off. Run after a meaningful unit of work, not after every reply.
liwm-self-improvementReview, replay and gate LIWM's own behavioural rule changes. Use for "LIWM review", or when deciding whether a proposed lesson should change future behaviour.
liwm-traceabilityAnswer "why did you do that?" from recorded evidence rather than plausible reconstruction. Use for "LIWM why", or whenever the user questions a choice you made.
liwmLatent Intent World Model - persistent understanding of how this specific user works. Consult before non-trivial work to load their learned preferences, decide how much to ask, and record what is learned. Handles "LIWM low/medium/high/auto/off/on", "LIWM profile", "LIWM why", "LIWM onboarding", "LIW