Skip to content

Glossary

A single place to look up every domain term Wearly uses. The glossary unifies vocabulary across the two layers of the project: the styling-agent prototype (the working app) and the Intelligent Book (its learning and reference companion). Definitions are short on purpose — each entry links to the chapter or rule file where the term is introduced in context.

Inspired by the glossary-generator skill in dmccreary/claude-skills, which the Level-2 intelligent-textbook framework expects.

A

Agent — A system defined by five properties Wearly's runtime exhibits: context-first stance, multi-step reasoning with a visible trail, tool use, memory across sessions, and an iterative reject-and-regenerate loop. See Chapter 1 · Product vision and Chapter 3 · Agent workflow.

Agentic loop — The reject-and-regenerate cycle: the user pushes back on a recommendation, the agent re-runs with the rejection recorded as a constraint, and the new reasoning trail surfaces what changed. See Chapter 3 § Reject & regenerate.

B

Blend note — A short, body-positive line that explains how two or more pieces work together (e.g., "the camel sweater warms the navy trouser"). Surfaced inside the reasoning trail.

Bloom level — Cognitive level a concept sits at: Remember → Understand → Apply → Analyze. Every node in the learning graph is tagged with one of these. See the learning graph.

Body-positive language contract — A runtime + documentation rule that forbids tokens like flaw, fix, hide, minimize, correct, problem area, slimming, shameful. Enforced in fit_tool.check_reasoning_for_forbidden_language() and in tests/test_documentation_integrity.py. See Chapter 6 · Fit-profile logic.

Body shape (preference) — A user-declared proportion preference (hourglass, pear, apple, inverted triangle, rectangle). Industry heuristic, not scientific taxonomy. Wearly never infers it from images. See evidence-and-references § Body-shape framing.

C

Calendar context — Upcoming event metadata read by Step 1 of the workflow. Drives occasion classification. Falls back to casual if no events are scheduled. See Chapter 7.

Citation chip — The small [occasion-rules#R8]-style tag appended to a reasoning line. Renders as a clickable chip in the UI; resolves via rule_refs.py to the named rule heading.

Color tier — One of +8 (great), +4 (okay), or −10 (avoid) — the three-value scoring scale Wearly uses for color-against-skin-tone judgments. See color-coordination-rules.md.

E

Evidence categories — The nine source categories every Wearly rule traces back to (recommender systems, outfit compatibility, color harmony, body-shape/fit, XAI, human-centered AI, wardrobe management, personalization, privacy). See evidence-and-references.

F

Favorite stores — A locally-stored list of retailers the user has saved. Personalizes the wardrobe-gap "where to look" suggestions without transmitting anything outside the device. See Chapter 8.

Fit profile — Optional, user-declared preferences: body_shape, modesty_preference, comfort_needs, style_goals, highlight_features, balance_areas. Every field is optional; defaults are safe. See Chapter 6.

Freshness score — A 0–1 value combining a frequency penalty and a recency penalty, with a 0.4 floor so a much-loved piece is never exiled from the rotation. See wear-history-rules.md.

G

Gap (wardrobe gap) — A required piece type the user does not own for the current occasion. Surfaced as a descriptive, never-promotional shopping suggestion. See Chapter 8.

H

Honesty contract — Four rules that govern every claim Wearly makes: styling is not exact science; body-positive language is non-negotiable; no fabricated citations; the user is the final reviewer. See evidence-and-references § The honesty contract.

K

Knowledge graph (runtime) — See Reasoning Graph. The earlier name "knowledge graph" was retired on display surfaces because McCreary's framework uses knowledge graph and concept graph interchangeably for the reader's concept DAG; we now call Wearly's runtime entity graph the Reasoning Graph to avoid the clash. File paths and the JSON source are unchanged.

Reasoning Graph — The wardrobe entities and typed relations the agent reasons over at runtime (User → CalendarEvent → OutfitRecommendation → WardrobeItem → ShoppingSuggestion / Feedback). Modeled in graph/graph.json with the schema in graph/schema.md. Distinct from the Learning Graph (the reader's concept DAG).

L

Learning graph — A directed acyclic graph (DAG) of the 28 concepts a reader needs to understand the system. Categories follow Bloom's progression. See the learning graph.

Level-2 intelligent textbook — Per Dan McCreary's framework, a book whose structure is driven by a concept-dependency graph and that ships MicroSims, glossary, FAQ, references, and Bloom-tagged concepts. Wearly is built to this level.

M

MicroSim — A small p5.js / vis-network interactive that lives in docs/sims/<name>/ alongside its index.md. Wearly currently ships four: color harmony, freshness, knowledge graph, learning graph.

O

Occasion tag — One of five canonical labels Wearly classifies any event into: work, gym, dinner, formal, casual. See occasion-rules.md.

P

Profile hash — A short hash of the merged fit profile, used by the UI to detect when the user has edited their profile mid-session and surface a "regenerate" banner. See Chapter 6.

Q

Qualified gap — A gap that survives a second check: the user owns nothing of the required type and the missing piece is plausibly shoppable for this occasion. Prevents noisy suggestions for niche events.

R

Regenerate banner — A small UI affordance shown when the merged profile hash changes, prompting the user to re-run the agent against the updated inputs. See Chapter 6.

Reject & regenerate — The agentic loop where the user rejects a recommendation and the agent re-runs with the rejection recorded as an exclusion. The reasoning trail shows what changed. See Chapter 3.

Required pieces — The minimum type-set an occasion needs (work needs top + bottom; gym needs activewear; formal needs a dress; etc.). See occasion-rules.md.

Result-dict contract — The single boundary between the agent and the UI: a Python dict with steps, recommendation, reasoning, gaps, color_score, weather, event, profile. See docs/architecture.md.

Rule citation (R\<N>) — An R1, R2, … heading inside a Skill rule file. Referenced from runtime reasoning via the slug <pack>#R<N> (e.g. occasion#R8). Registry: rule_refs.py.

S

Seven-step workflow — Wearly's canonical reasoning order: determine occasion → load profile → check weather → filter wardrobe → build outfit → check gaps → score color. See Chapter 3.

Skill package — skills/wearly-styling-agent/ — eight rule files plus a SKILL.md that follows the agentskills.io specification.

Skin-tone palette — One of three named palettes (warm olive, cool fair, deep warm) that Wearly's color-score formula reads from. See color-coordination-rules.md.

T

Tradeoff record — A short structured note attached to a recommendation when the agent made a non-obvious choice (e.g., picked a slightly less fresh item because the color score was much higher). Surfaced in the reasoning trail.

W

Wardrobe filtering — Step 4 of the workflow: narrow the full closet to the candidate pool that matches occasion tags + season + availability + prior rejection exclusions. See wardrobe-filtering-rules.md.

Wardrobe gap — See Gap.

Wear history — Per-item worn_count and last_worn_date, written when the user confirms an outfit. Feeds the freshness score. See wear-history-rules.md.

Weather context — Live Open-Meteo data, with a seasonal fallback when the user is offline. Drives layering thresholds and season tags. See Chapter 7.

Wishlist — A locally-stored list of items the user wants to acquire next, with optional priority and linked_gap fields. Used by infer_wishlist_taste() to bias selection toward acquired-taste signals.

Pattern follows glossary-generator from dmccreary/claude-skills. Add a term when it's used in at least two places (a chapter and a rule, or two chapters) and a reader can't infer its meaning from context.