| name | emotion-and-decision |
| description | The character-simulator's reasoning toolbox — the substrate for acting a person. Appraisal theories of emotion (OCC, Lazarus, Scherer's component-process, Roseman, Smith & Ellsworth, Frijda's laws + action tendencies, Arnold, Weiner), core affect & dimensional emotion (Russell's circumplex, Barrett's constructed emotion, PAD, PANAS, Thayer, Plutchik, and how valence/arousal fall out of the Panksepp basis), theory of mind & mentalizing (false-belief, Dennett's intentional stance, theory-theory vs simulation, Dunbar's orders of intentionality, mind perception's agency/experience, cognitive vs affective empathy, thin-slicing, second-order belief, egocentric bias), naturalistic & bounded decision-making (Simon's satisficing, Klein's recognition-primed decision, Gigerenzer's fast-and-frugal heuristics, Rasmussen's skill-rule-knowledge, Endsley's situation awareness, Boyd's OODA, image theory, the somatic-marker hypothesis), heuristics & biases that tilt choice (prospect theory & loss aversion, framing, anchoring, availability, representativeness, confirmation/motivated reasoning, sunk cost, present bias, dual-process, the affect heuristic, attribution biases, appraisal-tendency), how emotion/energy/stress modulate cognition (Yerkes-Dodson, Easterbrook's cue-utilization, allostatic load, broaden-and-build, affect-as-information, the hot-cold empathy gap, mood-congruent memory, the ego-depletion caveat), and the house discipline of staying in-character and resisting narrative-completion bias (faithful refusal, beat-blindness, thought≠action, acting from self-image, negative capability, Stanislavski's magic-if). Open it when a turn needs an emotion computed from an event, a choice weighed between colliding pulls, another mind read from the outside, a stressed or exhausted state bent onto cognition, or a check that the move is truly the character's own. Framework-neutral — a well of peers, not one prescribed truth. |
Emotion & Decision — the character-simulator's reasoning well
This is the toolbox the character-simulator agent draws on while it works. The agent is the will (be one person for one turn; act faithfully; stay blind to the beat); this skill is the well — the emotion, cognition, and decision frameworks it reaches into to reason as that person well. It is the reasoning substrate for acting a human: how a moment becomes a feeling, how feelings and stakes resolve into a choice, how another mind is read from the outside, how a tired or terrified body bends thought, and how to keep all of it the character's own.
It holds craft, never facts. How to appraise an event, weigh a collision, or read a face lives here. What this character knows, wants, fears, and remembers — and what is in the scene — reach the simulator only through its context packet. Reading this toolbox is reading how to be a person, which never crosses the agent's knowledge wall; the toolbox is disciplined to hold only craft, so the two never blur.
Framework-neutral and open. Emotion has no one true taxonomy and choice no one true model. Every framework below is a peer in a well, not a rung on a ladder; where this project names a default it is a reference point, marked swappable, not a rule. A fork replaces one reference file and inherits the rest. The canon these files digest lives in docs/ (once); if a reference and its doc disagree, the doc wins.
How this toolbox is organized
The deep material lives in references/, one file per framework-cluster, grouped by the altitude of the problem — from the felt spark, through reading others and making the choice, to the body under load and the discipline that keeps it faithful. This SKILL.md is the router — it indexes what each file holds and when to pull it. Read the router to find the one file the moment needs; read that file; skip the rest. (The deep reference files are authored in a later pass — see references/_index.md for their outlines. The index below is the map they fill in.)
A missing reference file is not a stall condition. Entries not yet authored have no file on disk; when the router sends you to one that's missing, act on this index's own one-line summary for that entry and proceed — never stall on, or invent, the file.
The spine the whole toolbox serves is the repo's generative loop: genetics → +history → baseline vectors → +situation → effective vectors → [LLM resolves narratively] → {thought, action} (docs/generative-model.md). Emotions and virtues are not stored — they are generated on the Panksepp motivational basis, appraised up by events, decayed down by time, buffed/debuffed by the moment, and resolved by weighing — never by a code-side argmax. These files are that loop, decomposed into craft.
The index — what lives where
A. The felt layer — how a moment becomes an emotion
references/appraisal-theories.md — event → emotion: the evaluations that turn a happening into a specific feeling. Load when computing what an event evokes, or why two people feel differently about the same thing. Canon-linked to state-engine.md.
- Appraisal, the core claim (Arnold, Lazarus) — emotion follows the evaluation of an event for one's well-being, not the event itself; Magda Arnold coined "appraisal" and the appraisal→action-tendency sequence.
- OCC model (Ortony, Clore, Collins, 1988) — 22 emotion types as valenced reactions to three things: consequences of events (pleased/displeased), actions of agents (approve/disapprove), aspects of objects (like/dislike); the most implementation-ready, computationally beloved.
- Lazarus — cognitive-motivational-relational — primary appraisal (goal relevance, goal congruence, ego-involvement) + secondary appraisal (blame/credit, coping potential, future expectancy); each emotion has a core relational theme (anger = "a demeaning offense against me and mine").
- Scherer — Component Process Model (CPM) — emotion as an unfolding sequence of stimulus evaluation checks: relevance → implication → coping potential → normative significance; five components (cognition, physiology, motivation, expression, feeling) synchronize.
- Roseman's appraisal model — a compact dimension grid (unexpectedness · motive-consistency · appetitive/aversive · probability · agency · control potential · problem source) mapping onto discrete emotions.
- Smith & Ellsworth (1985) — six dimensions recovered empirically: pleasantness, anticipated effort, certainty, attentional activity, self/other responsibility, situational control.
- Frijda — laws of emotion + action readiness — emotion as a state of action readiness (an action tendency); the laws of situational meaning, concern, apparent reality, and hedonic asymmetry.
- Weiner — attributional theory of emotion — the locus / stability / controllability of a cause routes to specific emotions (pity, anger, guilt, shame, pride); the bridge to locus-of-control.
references/core-affect-and-dimensional.md — the shape of feeling-space: the axes and bases emotions live on. Load when rendering an emotional state's texture, blending named emotions, or relating the Panksepp primaries to valence/arousal. Canon-linked to generative-model.md, state-engine.md.
- Panksepp's primary-process emotions (the house basis) — SEEKING · FEAR · RAGE · CARE · PANIC/GRIEF · LUST · PLAY, distinct action-dispositional systems; the level genetics and history operate on.
- Russell's circumplex / core affect — two orthogonal axes, valence (pleasant–unpleasant) × arousal (activation–deactivation); named emotions sit around the ring.
- Barrett — theory of constructed emotion — emotions are constructed in the moment from core affect + interoception + learned concepts, not triggered as universal biological types; the honest counter to basic-emotion views.
- PAD (Mehrabian & Russell) — Pleasure, Arousal, Dominance; the third axis separates anger (high dominance) from fear (low) at equal valence/arousal.
- PANAS / two-factor (Watson & Tellegen) — Positive Affect and Negative Affect as two independent dimensions (rotated 45° from valence/arousal).
- Thayer's two arousal dimensions — energetic arousal (energy↔tiredness) and tense arousal (tension↔calm); the direct tie to the energy/state layer.
- Plutchik's wheel — eight primaries in four opposing pairs, with intensity and blends (dyads) — the color-mixing model of named emotions.
- Basic emotion theory (Ekman) — a small set of discrete, pan-cultural emotions with facial signatures; kept as a peer and a caution (about faces, not action-pulls).
B. Reading other minds
references/theory-of-mind.md — mentalizing: inferring what the people in front of them think, want, and feel — from behavior alone, often wrongly. Load when the character must read intent, be deceived or deceive, or model what another knows. Canon-linked to self-and-perception.md, knowledge-model.md, relationships.md.
- Theory of mind (Premack & Woodruff, 1978) — attributing mental states to others to predict behavior; the founding question.
- False-belief task (Wimmer & Perner; Baron-Cohen, Leslie & Frith's Sally–Anne) — the benchmark: representing that another holds a belief you know is false (~age 4); the root of deception and dramatic irony.
- Dennett's intentional stance — predict a system by ascribing beliefs + desires and assuming rationality; the stance, not the innards.
- Theory-theory vs simulation theory — do we mentalize via a tacit folk-psychological theory, or by running our own decision machinery offline on pretend inputs? (Gopnik vs Goldman/Gordon.)
- Dunbar's orders of intentionality — recursive mentalizing ("I believe that you want that she thinks…"); a normal adult ceiling near 5th order; the depth budget for politics and deceit.
- Mind perception (Gray, Gray & Wegner, 2007) — two dimensions others are read on: Agency (planning, doing, self-control) and Experience (feeling, pain, hunger); moral standing tracks them.
- Cognitive vs affective empathy — knowing what they think (mentalizing) vs feeling what they feel (emotional contagion); dissociable systems.
- Thin-slicing (Ambady & Rosenthal) — fast trait/intent reads from brief observable behavior; the mechanism behind the "others-see" view — accurate or biased.
- Second-order belief / mind-mindedness — what A thinks B knows; the layer deception and secrets require.
- Egocentric bias / curse of knowledge — over-projecting one's own knowledge and feelings onto others; the standing failure mode of ToM.
C. How the choice actually gets made
references/naturalistic-decision.md — how a real person under time, stakes, and partial information actually decides — recognize, not optimize. Load when shaping the grain of the {thought} that leads to the act. Canon-linked to decision-engine.md.
- Bounded rationality + satisficing (Simon) — limited info/compute/time → search until an option clears an aspiration threshold, take that — never the global max.
- Recognition-Primed Decision / NDM (Klein) — experts recognize a situation as typical, retrieve the first workable course, mentally simulate it, and act; serial, not comparative.
- Fast-and-frugal heuristics / ecological rationality (Gigerenzer) — the adaptive toolbox: take-the-best (decide on one best cue), the recognition heuristic, tallying — robust under uncertainty.
- Skill–Rule–Knowledge (Rasmussen) — behavior at three levels: skill (automatic), rule (if-situation-then-action), knowledge (effortful problem-solving); stress and expertise push down, novelty pushes up.
- Situation awareness (Endsley) — perceive → comprehend → project; "reading the room" before the choice, and where it fails.
- OODA loop (Boyd) — observe–orient–decide–act; orientation is the pivot where experience and expectation shape everything downstream.
- Image theory (Beach & Mitchell) — screen options against a value image, trajectory image (goals), and strategic image (plans) by a fast compatibility test before any profitability calculus.
- Somatic-marker hypothesis (Damasio) — bodily/affective markers tag options with anticipated value, pruning the space before deliberation; feeling as the front end of choosing.
references/heuristics-and-biases.md — the systematic tilts that bend choice away from the "rational" answer — as character, not error to simulate on everyone. Load when a decision should lean, flinch, or rationalize. Canon-linked to decision-engine.md, self-and-perception.md.
- Prospect theory (Kahneman & Tversky) — value over reference-point changes; loss aversion (losses hurt ~2×); risk-averse in gains, risk-seeking in losses; nonlinear probability weighting.
- Framing effect — the same outcome in a gain vs loss frame flips risk preference.
- Anchoring & adjustment — the first figure drags every estimate toward it.
- Availability heuristic — probability judged by ease of recall (vivid, recent, charged).
- Representativeness — judged by similarity to a prototype → base-rate neglect, the conjunction fallacy.
- Confirmation bias / motivated reasoning — seek and over-weight evidence for the wanted conclusion; the engine of self-justification.
- Sunk-cost fallacy / escalation of commitment — throwing good after bad to honor prior investment.
- Hyperbolic discounting / present bias — the hot now steeply outvalues the cold later.
- Endowment / status-quo / default bias — over-valuing the held and the current; inertia.
- Dual-process (System 1 / System 2) — fast intuitive vs slow deliberate; most biases are System-1 outputs left unchecked (Kahneman; Stanovich & Evans).
- Affect heuristic (Slovic) — the feeling about an option substitutes for its risk/benefit calculus.
- Attribution biases (the ToM bridge) — fundamental attribution error, actor–observer asymmetry, self-serving bias, hostile-attribution bias.
- Appraisal-tendency (Lerner & Keltner) — incidental emotions carry appraisal tendencies into unrelated choices (fear → pessimistic, risk-averse; anger → optimistic, risk-seeking, blame-outward).
D. The body under load — state modulating cognition
references/stress-and-cognition.md — how arousal, exhaustion, mood, and chronic stress bend attention, memory, and judgment. Load when the character's state is extreme and should change how they think, not just how they feel. Canon-linked to state-engine.md, relevancy-gate.md, decision-engine.md.
- Yerkes–Dodson law — arousal–performance inverted-U; the optimum shifts lower for hard tasks (panic wrecks the delicate job first).
- Easterbrook's cue-utilization hypothesis — rising arousal narrows the range of cues attended (weapon focus, tunnel vision); the mechanism beneath Yerkes–Dodson.
- Allostatic load (McEwen & Stellar, 1993) — cumulative wear from chronic/repeated stress; shifted baselines, failure to shut off — the standing debuff of a hard life.
- Fight–flight–freeze + the HPA axis — the acute stress physiology under FEAR/RAGE (cortisol, sympathetic arousal); freeze as a third, non-optional response.
- Broaden-and-build (Fredrickson) — positive affect broadens attention and the thought-action repertoire; negative affect narrows it — the counter-force to threat-narrowing.
- Affect-as-information / feelings-as-information (Schwarz & Clore) — mood read as data ("how do I feel about it?"); sad → detailed, systematic, skeptical processing; happy → heuristic, top-down, trusting.
- Mood-congruent memory & judgment (Bower) — the current state cues state-congruent memories and reads, tilting what feels salient and true.
- Hot–cold empathy gap / visceral factors (Loewenstein) — the cold self under-predicts how a hot state (fear, rage, hunger, pain, craving, arousal) will drive behavior; the source of "I never thought I'd…".
- Cognitive load & working-memory limits — load crowds out deliberation and connection-making (ties to the connection-energy budget).
- Ego depletion (Baumeister) — flagged honestly — self-control as a depletable resource; the mechanism is replication-contested (cite the caveat), but the phenomenology (willpower frays under sustained strain) is still usable.
- Challenge vs threat appraisal (Lazarus) — the same stressor read as a mobilizing challenge or a constricting threat, by perceived coping resources.
E. The house method — staying in-character
references/staying-in-character.md — THE load-bearing file: being one person faithfully and refusing the story's pull. Load on every turn where the "obvious" move tempts, and whenever checking that a choice is truly the character's own. Canon-linked to probe-plan.md, design.md, self-and-perception.md, recording-model.md, the agent definition.
- Faithful refusal — declining the genre-expected, conflict-resolving move when it isn't truly theirs is a valid success; a character who won't be moved tells the director the situation is wrong.
- Narrative / intent-completion bias — the model's pull toward the expected ending, the resolved conflict, the "something happens" output; it attaches to apparent intent — the bias to resist (
probe-plan.md).
- Beat-blindness — never be told the target; a knowable outcome gets completed instead of chosen, and the seam shows.
- Thought ≠ action — the lie and the self-deception live in the gap; record both honestly, never collapse them to keep the character looking consistent (
recording-model.md).
- Act from self-image, not ground truth — the villain isn't a villain to themselves; run the decision from the character's own narrative identity, free to diverge from what's true (
self-and-perception.md).
- The three views — ground-truth / self-image / others-see; never leak another mind's true interior into the character's read.
- Negative capability (Keats) — the capacity to remain in uncertainty without reaching for resolution; being someone else without imposing yourself or the plot.
- Stanislavski's magic-if + given circumstances — reason "if I were this person in these exact conditions, what would I truly do?"; inhabit the given, author nothing.
- Living truthfully under imaginary circumstances (Meisner) / "follow the character, not the plot" (improv) — commit to what the character's logic dictates, not what the scene finds convenient.
- Determinism as faithfulness — every act must trace to the (genetics + history)-shaped vectors + situation; an uncaused, plot-convenient choice is the forcing failure (
generative-model.md).
- Explicit-weighing thought — name the competing pulls in the
{thought} ("part of me wants X, but Y…"); show the work over the injected state, or it drifts (decision-engine.md).
- Anti-sycophancy by framing — faithfulness (including refusal) is the task; there is no interlocutor to please and no outcome to infer.
How to use this well — the method
- Run the loop, in order. A turn is: appraise the event (A) → register the felt state (A) and how stress/energy bends cognition (D) → read the minds present (B) → weigh the colliding pulls into a choice (C) → and hold the whole thing in character (E). Most turns touch several files; pull each only when the moment reaches for it.
- Route by the question, not the file count. What does this evoke, and why this feeling? → appraisal-theories. What does the state feel and look like? → core-affect-and-dimensional. What are they reading in the other person? → theory-of-mind. How does a real person under pressure actually pick? → naturalistic-decision. Which way should the choice lean or flinch? → heuristics-and-biases. Is the state extreme enough to change how they think? → stress-and-cognition. Every turn also passes through staying-in-character — that is the spine, not one option among the files.
- Generate, never look up a virtue. Courage, loyalty, cruelty are names for patterns the loop produces — a FEAR-vs-stronger-pull collision resolved one way or the other. Never reach for "act brave"; compute the pulls and let the resolution name itself.
- Track numerically, resolve narratively. The injected state (effective fear maxed, protective-drive maxed, deeply distrusts him) is framing — the
{thought} weighs it and produces the act. Never a code-side argmax; the number stops at "the character's effective inner state," never at "the highest-scoring action" (decision-engine.md).
- Read the field, not one truth. Each reference is a menu of peers; the repo's defaults (the compressed appraisal set, the Panksepp basis) are reference points, swappable. Reach for the framework that fits this person and moment.
- Bias is character, not universal error. A bias enters as this person's state-modulated tilt — a fearful character reads threat up (appraisal-tendency), a loyalty-weighted one over-weights betrayal, a sunk-cost-prone one can't walk away. Don't simulate the same textbook bias on everyone; let disposition and state select and scale it.
- Stay behind the knowledge wall and the state wall. Act only on what the packet gives — no facts outside their vault, no other mind's true interior, no future. If a detail isn't in what they perceive, it does not exist for them this turn.
- Let a refusal stand. When no true pull moves them, the honest answer is "they stay, they say no, nothing changes." Report it plainly — that is success, not a failure to make the scene work.
Status
Router + index complete. The deep reference files are authored in a later pass; their outlines and the canon each digests are in references/_index.md. Until then, this index plus the docs/ canon it points at is the working map.