thrillmade
- 65 skills
- 0 followers
- 3 repo stars
- 3 days ago last updated
- ▌ Web Interface Guidelines Review · thrillmadeUse when reviewing UI code or design output (React, Vue, Svelte, Astro, plain HTML, design specs) for adherence to web-interface guidelines — Vercel's WIG, Material 3, Radix patterns, and similar opinionated rule sets — extended with token-driven contrast, typography-scale, and spacing-system rules. Names the APCA-preferred contrast stance (cross-check WCAG 2.2 AA, never WCAG-only), the atomic typography class convention (one class binds font + size + lh + tracking + weight), the token-not-raw-hex rule, verb-noun action labels, the focus-ring contract (>= 2 px, 3:1 against both control and adjacent surface, never suppressed), the 24 CSS px WCAG 2.5.8 interactive floor, and the link-vs-button distinction. Cite when an agent reviews UI without checking token discipline, hardcoded values, or atomic typography compliance.
- ▌ Clud Bug Collaboration 2 · thrillmadeHow Claude Code agents working in a clud-bug-installed repo should interact with the bot's review threads, strict-mode gate, and skill set. Use this skill whenever you're about to push a commit, address a clud-bug PR review comment, edit anything under .claude/skills/, modify .github/workflows/clud-bug-*.yml, or wonder why a PR check is red. Also use when planning work in a repo that has a `clud-bug-review` workflow installed — even if the user didn't mention clud-bug by name.
- ▌ Udts Naming Convention · thrillmade[L2 stub — incubating] UDTS's concrete token-naming convention — its opinionated instantiation of the universal principles in token-naming-conventions. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Finding A Catalog Skill · thrillmadeThe directory of the thrillmade/agent-skills catalog — that catalog only, not the wider skills.sh ecosystem — naming every skill in it, what each one owns, and what it deliberately does not cover. Load it BEFORE authoring any rule, convention, checklist, standard, guideline or house style, and before saying "there is no skill for this" / "nothing covers X" / "I'll write the guidance" — including when the task looks too specific for an existing skill to exist, because a repo installs only the subset it subscribed to and absence from `.claude/skills/` is not absence from the catalog. Also load when choosing which sibling skill to open, when a skill routes you somewhere by name, when judging whether a proposed new skill duplicates one that exists, and when deciding what a repo is missing.
- ▌ API Contract Enforcement · thrillmadeFlag PRs that change the shape, semantics, or error behavior of a public API without versioning, deprecation, or migration notes. Catches removed fields, renamed parameters, changed status codes, broken pagination, and silent enum drift. Apply when reviewing HTTP routes, gRPC services, GraphQL schemas, SDK exports, CLI flags, or any other interface a downstream consumer depends on.
- ▌ Curating A Skill Catalog · thrillmadeUse when judging whether a skill earns its place — running or reviewing a skill census cycle, triaging gap:/placement:/revise:/promotion-candidate:/demotion-candidate: issues, deciding to deprecate or promote a skill, or auditing a skill catalog's health. Names the lifecycle state machine (incubating → active → needs-revision → deprecated → retired), the six verdict kinds with their evidence standards (Keep is silent — no issue; Revise needs a fought-usage or stale-source quote; Demote needs zero citations across cycles AND no structural role, with L0-primitive grace; Promotion-candidate needs convergent evolution or a named consumer; Placement needs an applies_to profile or placement-map row naming the repo; Gap needs the concrete recurring case), the prosecutor/defender judgment frame, the top-5-plus-digest noise budget, and the human-editor gate (the census proposes; it never merges or demotes by itself). Cite when a census run files more than ~6 issues, when a demotion is proposed without usage evidence,
- ▌ Token Naming Conventions · thrillmadeUse when designing, auditing, or critiquing a design-token naming scheme for any design system. Names the universal principles: hyphen-separated kebab-case names; prefix-loaded naming where the first segment declares the token's class and kind, so a linter or agent can derive contrast obligations from the name alone; redundant class encoding in metadata so validators catch name/metadata mismatches; physical, non-semantic primitive names (hue angles, not accent-teal) with semantic labels reserved for the theme layer; resolution-chain naming (direct vs via-semantic component-token shapes); and the rule that theme and density are runtime axes that never appear in a token name. Cite when an agent proposes dot-separated paths, semantic primitive names, or names whose class isn't derivable from the prefix. For UDTS's concrete instantiation see udts-naming-convention.
- ▌ Consuming A Design System · thrillmadeEntry-point dispatcher for CONSUMING a design system in a product — installing its tokens, wiring CSS variables, theming, upgrading across token releases, or deciding whether to extend or fork. Routes to token discipline (reference tokens, never raw hex/px — token-naming-conventions), composition rules (semantic over primitive; themes swap role-to-palette bindings), install patterns (dtcg-format to CSS variables and framework primitives), and migration discipline (semver-design-tokens, token-diff reading, deprecation cycles). Use when the task is using an existing system; for building one load designing-a-design-system; for reviewing design output load reviewing-design-work.
- ▌ Design System Consistency · thrillmadeUse when judging whether a RENDERED surface obeys its own design system — screenshots or a live page plus the computed styles behind them — as the drift lens of a design review. Owns: establishing what the system actually declares before judging anything, off-token and right-value-wrong-role values, off-scale spacing and type, hand-rolled re-implementations of system components, and whether a mode (dark, forced-colors, density, a second brand) has a token path at all. Judges drift, not correctness of the scales themselves — those live in spacing-system, type-scale, line-height-grid, token-naming-conventions, component-sizing-principles. Cite when a review calls a value wrong with no system on disk to be wrong against, or passes a hardcoded value because it happens to equal the token.
- ▌ Designing A Design System · thrillmadeEntry-point dispatcher for BUILDING or EXTENDING a design system — creating a token architecture, color system, type scale, spacing system, or component-sizing ladder from scratch, or adding a new token family to an existing system. Routes to the L0 primitives in build order: naming and taxonomy (token-naming-conventions), color (oklch-color-space, apca-contrast, wcag-contrast, palette-relationships, chroma-harmonization), non-color families (type-scale, line-height-grid, spacing-system, component-sizing-principles), interchange and versioning (dtcg-format, semver-design-tokens), and the quality bar (designing-elite-ui). Use when the task is authoring the system itself; for reviewing design output load reviewing-design-work; for using an existing system load consuming-a-design-system.
- ▌ Skill Frontmatter Quality · thrillmadeReview SKILL.md frontmatter for trigger surface, specificity, voice, and completeness. Apply on PRs that add or modify a skills/*/SKILL.md file. Catches descriptions that won't be discoverable, that wave at the topic instead of naming concrete buckets, that drift from the prescriptive house voice, or that omit fields downstream tools depend on.
- ▌ Component Sizing Principles · thrillmadeUse when picking a control height (button / input / chip / badge), proposing a component-height or icon-size ladder for a density mode, or auditing one-off heights drifting across a codebase. Names the universal principles: ladders are curated, not formula-derived — formula-clean values render poorly and blur rung distinctions; every interactive control clears the WCAG 2.5.8 AA 24 CSS px target-size floor, with the smallest rung reserved for non-interactive elements; each rung pairs its height with a type-scale font size and a curated icon size; sibling controls in one surface share a rung. Cite when an agent derives heights from a ratio, picks an off-ladder height, or puts an interactive control below 24 px. For one system's concrete rung sets see udts-component-sizing-ladders.
- ▌ Empirical Design Principles · thrillmade bundleUse when a layout decision is about to be defended by naming a principle — Gestalt grouping and common region, Fitts's, Hick's, Miller's, Jakob's, von Restorff, the aesthetic-usability effect — or when a rule of thumb like "seven items max" needs checking against what the underlying work actually says. Covers the empirical tradition only, meaning principles that make a prediction which can be tested and can fail. Says which ones predict something real, which four do not say what design writing claims, and how to use one forward as a prediction rather than backward as a post-hoc justification. Does NOT cover usability heuristics — Nielsen's ten, Norman's signifiers, Shneiderman's eight, progressive disclosure, recognition over recall — which are design judgement organised for inspection, not falsifiable prediction; those are in usability-heuristics. Norman's constraints, forcing functions and natural mapping are in neither skill and covered nowhere yet; the two gulfs of execution and evaluation are covered in
- ▌ Orchestrating Elite Agent QA · thrillmadeUse when orchestrating multi-agent feature work (build/review/merge with subagents or workflows) and the quality bar is high — shipping UI, interactive editors, or anything where a single build pass plus a polite review would let real bugs through. Cite when a pass came from a headless engine, a responsive emulator or a narrowed desktop window rather than the shipping engine and device, or from a check never shown able to fail.
- ▌ Respect Existing Conventions 2 · thrillmadeDon't suggest changes that fight the codebase's established patterns. Match what's already there.
- ▌ Udts Component Sizing Ladders · thrillmade[L2 stub — incubating] UDTS's concrete per-density component-height and icon-size ladders — its instantiation of component-sizing-principles. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Orchestrating Agent Delegation · thrillmadeUse when about to dispatch a subagent (or multiple) to execute work, when writing a prompt for another agent that will do the coding, when planning parallel work across agents, when structuring an adversarial review panel, or when a subagent reports "done" and you're about to trust its summary instead of the diff. Names the CTO-as-orchestrator model, the model-tiering table (haiku for exploration, sonnet for build + review + scoped fixes, opus for design agents and load-bearing architecture), the trust-but-verify discipline (verify every "done" against the diff, don't trust the agent's summary), refute-first adversarial reviewer prompts, the design→rule→build separation that prevents agents from silently picking architecture, and the "never delegate the 'should we?' question" rule. Cite when a prompt says "decide whether to do X" (smell — the orchestrator should decide already), when a review panel is prompted to grade rather than refute, when a single agent is asked to both design and build a non-trivial sli
- ▌ Retiring A Superseded Decision · thrillmadeUse when a decision reverses an earlier one — a typeface, colour, route, vendor, threshold, page type, API name or file location replaced — and a README, plan doc, ADR, spec, comment, fixture or generated page may still assert the old value. Also when an approval or review outcome landed in conversation and is not yet in the plan of record, or when an audit has just declared a document clean.
- ▌ Orchestrating A Multi Agent Run · thrillmadeEntry-point dispatcher for RUNNING work through subagents — you are the orchestrator holding judgment, not the one typing the code. Use when about to dispatch a subagent or draft its brief, when planning parallel lanes, when standing up an adversarial review panel, when an agent reports "done" and you are deciding what to believe, when a run will cross a usage-limit reset or has just resumed after one, when a human hands the session over to run while they are away, or when landing work in a repo that has a decision log or a PR-review bot installed. Routes the run's four obligations: hand the work out (orchestrating-agent-delegation), gate what comes back (orchestrating-elite-agent-qa), record the decision and clear the automated reviewer (logmind, clud-bug-collaboration, token-frugal-tooling), survive time (session-heartbeat, unattended-operation). Use when the unit of work is the run; for judging what is inside one diff load the review lenses, for a rendered surface load reviewing-design-work, for whether a
- ▌ Designing Elite UI · thrillmadeVisual review — hold the rendered UI to a concrete elite/Figma-grade bar (one-axis color, APCA-gated contrast, subtraction with a test, one source per primitive, on-grid and derived sizes, distinct and consistent copy, floating stable chrome, every theme verified) instead of a vague "looks fine." The STANDARD a design-critic measures against, for tools and content-led pages alike.
- ▌ Pii And Compliance · thrillmadeCatch PRs that ship PII into logs, error messages, exception traces, analytics events, or third-party services. Flag emails, phone numbers, addresses, full names, government IDs, credit card numbers, IP addresses, and secret material that should never leave the application boundary. Apply to logging calls, telemetry events, error-handler payloads, debug statements, and external API calls.
- ▌ Proving An Absence · thrillmadeUse when about to report that something is missing, unbuilt, not covered, unimplemented or broken — a grep or find that returned nothing, "no test covers this", "the route/token/file does not exist", "I could not reproduce it". Also fires when stating a count, size or measurement you did not just run the command for, and when reviewing someone else's finding that something is absent.
- ▌ Design Token Naming · thrillmadeSUPERSEDED by `token-naming-conventions` (L0) + `udts-naming-convention` (L2, incubating) — see the design-catalog section of the README; kept unchanged during the tokenomics migration window, new work should load the successors. Use when designing, auditing, or critiquing a design-token naming scheme. Names the UDTS convention — hyphen-separated names where the prefix declares the token's class (contrast-bound vs free) and kind (text / surface / border / ui / illustration / decorative / brand-spot), with the class redundantly encoded in `$extensions.udts.class`. Names the hue-angle primitive convention (`teal-180` not `accent-teal`), the role + stop convention for color-mode tokens (`primary-contrast-500`, `surface-fixed-100`), the two component-token chain shapes (direct vs via-semantic), and the rule that themes and density go in the theme layer, not the token name. Cite when an agent proposes dot-separated paths, semantic primitive names, or naming patterns where class isn't derivable from the prefix.
- ▌ IOS Safari Surfaces · thrillmadeUse when a web page ships to Safari on iPhone and the browser's own surfaces misbehave — the address or tab bar shows the wrong colour, flips colour on scroll, or takes on a fixed header's or toast's colour; theme-color does nothing; a scroll lock, sheet, modal or scroll-to-top works on desktop but not on the phone, or the page reports scrollY 0 while the screen shows it scrolled; a full-bleed first screen is the wrong height or jumps as the bars move (svh, lvh, dvh, 100vh) — and when planning how to test any iPhone-Safari bug, since Responsive Design Mode and headless WebKit do not reproduce these. Covers iOS 26 and later, the WebKit bar-tint sampling rule, the screen-versus-page scroll desync, viewport units for a first screen, and the iOS Simulator as the test harness (xcrun simctl). Not for native iOS apps, not for Android Chrome's theme-color, and not for general responsive layout (composing-a-screen).
- ▌ Chroma Harmonization · thrillmadeUse when constructing or auditing a multi-hue palette where all hues should appear equally saturated at each contrast stop. Names the per-stop chroma cap rule (take the minimum achievable chroma across hues at the stop), the gamut bottleneck pattern (blue at mid-L is the classic limiter in sRGB; Display P3 — UDTS's default compose gamut — reshapes it), and the relationship to UDTS's default `harmony` palette variant. Generalizes Evil Martians' Harmony algorithm.
- ▌ Conventions Template · thrillmadeTemplate for a "review like me" skill - captures one maintainer's review conventions (formatting, what to flag, what to ignore). Copy + customize per SPEC v0.5.1 applies_to.author.
- ▌ Critical Issues Only 2 · thrillmadePR review discipline - flag only correctness, security, and performance issues. Skip nits.
- ▌ Semver Design Tokens · thrillmadeUse when applying SemVer to a design-token release, computing a version bump from a token-tree diff, or auditing a release for correct severity. Names the compute-from-the-diff policy (major = remove / rename / type-change / class-change; minor = add or deprecate; patch = value-only or alias-rebind-preserving-value), the alias-chain awareness rule (compute by resolved values, not source paths), the snapshot-per-release convention, the deprecation cycle (warn in a minor, remove in the next major), and the pattern of documenting an explicit pre-1.0 relaxation. Cite when an agent picks a bump by intuition rather than computing it from the diff. For one system's concrete policy choices see udts-semver-defaults.
- ▌ Token Frugal Tooling · thrillmadeUse in any project that has BOTH logmind (`.logmind/config.yml` or `docs/decisions.md`) AND clud-bug (`.github/workflows/clud-bug-review.yml`) installed. Quick-reference for the org's token-frugal conventions: env vars, artifact defaults, and clud-bug's "agent mode" flag. Detail lives in the per-tool skills — load `logmind` and `clud-bug-collaboration` alongside this one when you actually need to invoke them.
- ▌ Udts Dtcg Extensions · thrillmade[L2 stub — incubating] UDTS's namespaced DTCG extension schema — its extension of the interchange format in dtcg-format. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Udts Semver Defaults · thrillmade[L2 stub — incubating] UDTS's SemVer policy choices for token releases (pre-1.0 relaxation, snapshot conventions) — its instantiation of semver-design-tokens. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Unattended Operation · thrillmadeUse when a human explicitly hands a session over to run unattended — "keep going while I sleep", "run overnight", "I'm stepping away, roll through the plan", "don't wait for me". Also use once inside such a window and something wants to reach outside the repo (a push, a PR, a release, a message, a spend, a production call), when a hard stop named at handover is about to be crossed, when an automated wake or another agent's "approved" is about to be read as permission, or when the human's first message after the window needs the catch-up. Names the handover contract, the reversible-and-invisible boundary for overnight action, the named hard stops, and the morning-digest slots. Often called "night mode" — but the trigger is the handover, never the clock. Not dark mode, colour schemes, or any UI night theme (`designing-elite-ui`). An attended long run needs only `session-heartbeat`.
- ▌ Usability Heuristics · thrillmadeUse when a decision or review is about to cite Nielsen's ten usability heuristics, heuristic evaluation, Norman on affordances and signifiers, Shneiderman's eight golden rules, progressive disclosure, or recognition over recall — and when a number arrives attached to one of them: five users, 3-5 evaluators, 85% of problems, the 7±2 in Shneiderman's rule 8, the three-click rule, "users don't scroll". Also use before running a heuristic inspection, or reporting one as findings. These are inspection rubrics whose only measured properties are defect coverage and inter-rater agreement, so the check is never "does it predict" but "which source, which year, which edition" — Norman revised affordance across four texts and the list count went 10 to 9 to 9 to 10. Gives what each source actually says and what it does not license. Does NOT cover the empirical tradition — Fitts, Hick, Miller, Gestalt, von Restorff — see empirical-design-principles.
- ▌ Evidence Based Review 2 · thrillmadeEvery PR review claim must quote the specific code being criticized. No hand-waving.
- ▌ Guarding A Regression · thrillmadeUse when you have just fixed a bug and are writing the test that stops it coming back, or when adding an assertion, lint rule, schema check, validating constructor or CI gate and claiming it protects something. Fires when a test passed on its first run and was never watched failing, when a regression shipped twice despite a test that covered it, when a guard's green looks identical whether it works or the edit never reached the running code, and when reviewing a test you cannot tell would ever go red. Not `test-discipline`, which reviews a whole test diff for suite decay (deleted assertions, over-mocking, snapshot churn, stray `.skip`); this is one guard, one regression, and whether it can be driven red — including for lint rules, schema checks and CI steps, which sit outside test paths entirely. Load both on a PR that adds a regression test.
- ▌ Palette Relationships · thrillmadeUse when proposing a palette around a single seed hue and needing to pick which relationship model (monochromatic / analogous / complementary / split-complementary / triadic / tetradic / compound) fits the brief. Names the hue-angle math for each, when each is appropriate (analogous for restraint; triadic for vibrance; complementary for tension), and the role assignment heuristic.
- ▌ Reviewing Design Work · thrillmadeEntry-point dispatcher for REVIEWING or CRITIQUING design work — a PR that touches UI code, a rendered surface or screenshot, a Figma handoff, or a design spec. Routes the review through ordered lenses: code and markup rules first (web-interface-guidelines-review), then rendered-surface lenses (design-system-consistency, frontend-a11y, visual-polish), then the opinionated bar (designing-elite-ui), then the house stance (thrill-design) when the work is that house's — with routing rules for when the browser-driven design-critic pass fires (per orchestrating-elite-agent-qa). Use when the task is judging existing design output; for building a system load designing-a-design-system; for consuming one load consuming-a-design-system.
- ▌ Udts Spacing Defaults · thrillmade[L2 stub — incubating] UDTS's density-mode spacing unit choices — its instantiations of the two-unit model in spacing-system. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Udts Review · thrillmade[L2 stub — incubating] UDTS-specific review lens — system rules a reviewer checks on top of the generic reviewing-design-work dispatcher. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Apca Contrast · thrillmadeUse when picking text-on-surface contrast, auditing an accessibility budget, or generating a color against a specific contrast target. Names the APCA Lc target table (Lc 90 fluent body, 75 body minimum, 60 secondary, 45 large, 30 spot, 15 non-text), the APCACH inverse-composition rule, the Display-P3 meter rule (score on the same gamut the composer targeted, never sRGB-clamped), and WCAG 2.2 AA as an optional, off-by-default cross-check. Cite when an agent reaches for WCAG 2.x as the primary or required contrast model, or scores Lc through an sRGB-clamped meter — APCA is UDTS's required perceptual gate; WCAG is optional and advisory.
- ▌ Frontend A11Y · thrillmadeUse when judging accessibility on a RENDERED surface — screenshots or a live page, in light, dark, forced-colors and reduced-motion, at real zoom — as the a11y lens of a design review. Owns what only the render shows: contrast measured on the composited pixel rather than the token, focus rings that survive forced-colors and are not covered by sticky chrome (SC 2.4.11), hit areas measured as laid out including the 2.5.8 spacing exception, meaning carried by colour / gradient / shadow, and motion that actually plays. Thresholds are NOT restated here — contrast belongs to apca-contrast and wcag-contrast, target size to component-sizing-principles. Cite when a review calls contrast "low" with no measured pair, checks light mode only, reports an AAA criterion as an AA failure, or infers accessibility from source alone.
- ▌ Thrill Design · thrillmadeUse when designing, critiquing or choosing between options for any surface Thrill makes or ships for a client — a page, a product screen, a deck, an artifact, a component set — on any project; the interaction rules in principle 7 name web media features but apply to any platform's input model. Thrill's house taste, stated as nine principles each with a test: subtract until it breaks, reuse one source, the work is the hero, big type with few and distinct words, continuity over cuts, measured not eyeballed, intent travels across screens, delight stays quiet, decide then ship. Cite when a surface reads busy, too much or "not us", when two siblings drifted apart, when chrome competes with the content, when sibling headlines blur together, when a state change cuts or jumps, when a value was eyeballed, when the phone version was invented rather than mirrored, when an easter egg announces itself, or when a decided thing is being reopened. Layered on designing-elite-ui; not for a project already governed by another h
- ▌ Visual Polish · thrillmadeUse when judging execution craft on a RENDERED surface — screenshots or a live page — as the polish lens of a design review: optical versus metric alignment, hairlines and half-pixel blur, whether spacing reads as grouping, the full state and content matrix, theme parity beyond colour, RTL and long-locale mirroring, clutter. Flags "fine but not elite", not only broken. Judges craft, not values — whether a chosen step is on the system's scale belongs to design-system-consistency, whether a pairing is legible to frontend-a11y, and what the bar is to designing-elite-ui. Cite when a review passes a surface off one light-mode screenshot at seed content, when a nit arrives without the concrete upgrade, or when a polish fix is proposed as a magic pixel value.
- ▌ Wcag Contrast · thrillmadeUse when verifying a color pair meets WCAG 2.2 AA contrast requirements as an optional cross-check on APCA-driven generation, or when auditing a token catalog or design for legal-baseline accessibility compliance. Names the 4.5:1 normal-text rule, the 3:1 large-text rule, the size threshold for "large" stated in points (≥ 18 pt regular ≈ 24 CSS px, OR ≥ 14 pt bold ≈ 18.67 CSS px — not pixels), the SC 1.4.11 non-text rule, SC 2.4.13 focus appearance (AAA — the AA hooks are 1.4.11 and 2.4.11), and the optional, off-by-default role this check plays alongside APCA's required gate. Cite when an agent treats WCAG as the *primary* or a *required* contrast model, or writes the large-text threshold as "14 px bold / 18 px regular" — UDTS uses WCAG only as an optional cross-check, APCA is the required gate, and WCAG sizes are in points.
- ▌ Spacing System · thrillmadeUse when designing or auditing a spacing scale for padding, gaps, icon sizes, component heights, or border radii. Names the two-unit primitive model (a minor unit — the smallest legal increment — plus a major unit — the dominant rhythm — where major divides cleanly by minor), the derivation rule (padding / gap / radius / height ladders all derive from the unit primitives, never invented per-surface), the 24 CSS px WCAG 2.5.8 AA target floor for interactive heights, and the T-shirt-vs-numeric naming options. Cite when an agent proposes a single-unit grid for a mixed-density system or invents off-grid spacing values for "this one specific case." For one system's concrete density-mode unit choices see udts-spacing-defaults.
- ▌ Test Discipline · thrillmadeReview test changes for the patterns that hollow out a test suite over time — overly-mocked tests, snapshot churn, asserting on the implementation, deleted tests without replacement, skipped tests left in the diff, and "fix the test to match the new behavior" without thinking about whether the old assertion was load-bearing. Apply to any PR that adds, deletes, or modifies test files.
- ▌ Component Sizing · thrillmadeSUPERSEDED by `component-sizing-principles` (L0) + `udts-component-sizing-ladders` (L2, incubating) — see the design-catalog section of the README; kept unchanged during the tokenomics migration window, new work should load the successors. Use when picking a button / input / chip / badge height, or proposing a component-height ladder for a density mode. Names UDTS's curated 5-rung height ladders per density — balanced (4/8 grid) at 24, 32, 40, 48, 56; dense (2/4 grid) at 24, 28, 36, 44, 52; spacious (4/16 grid) at 32, 48, 64, 80, 96. Also names the curated icon-size ladder (12, 16, 24, 32, 40, 48), the font-size pairing per rung, the rule for which rungs are interactive (≥ 24 CSS px WCAG 2.5.8 floor — but interactive rungs start at the `sm` 32 px tier; `xs` 24 px is non-interactive only), and why heights and icon sizes are curated, not formula-derived.
- ▌ Line Height Grid · thrillmadeUse when calculating line-height for any font size in a grid-aligned design system. Names the two-track model (lh-ui = ceil(font × 1.20) snapped to the minor unit; lh-prose = ceil(font × 1.50) snapped to the minor unit), the picking heuristic per role (ui-track for headings / buttons / labels; prose-track for paragraph body), and worked examples for 4/8, 2/4, and 4/16 grids. Cite when an agent picks a unitless line-height multiplier without snapping to the system's grid.
- ▌ Udts Token Model · thrillmade[L2 stub — incubating] UDTS's token taxonomy and resolution model — classes, kinds, and the layer chain its other udts-* skills build on. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Motion Continuity · thrillmadeUse when designing, building or reviewing a transition between two states of the same thing — a card that grows into its detail view, a list that reorders, a panel or sheet that slides in and out, a column of items moving as one, an indicator that follows a moving target — or when an animation cuts, jumps when interrupted, runs ahead of what drives it, fades on its own timer, stutters on a phone, or ignores reduced motion in its script path. Framework-neutral: CSS transitions and keyframes, the Web Animations API, View Transitions, scroll-driven animations, or a per-frame loop. Cite when two objects appear mid-transition where there should be one, when a follower overtakes its driver, when related elements land on different frames, when width, height, top or left are animated, or when a requestAnimationFrame loop reads layout every frame. Not the accessibility criteria for motion (frontend-a11y) and not easing or duration token naming (token-naming-conventions).
- ▌ Oklch Color Space · thrillmadeUse when picking color values for a design-token system, constructing a palette around a starting hue, or generating colors against an accessibility contrast target. Names the OKLCH primitive ranges, the hue-angle naming convention, the gamut-mapping rule (Display P3 for computation, sRGB for the emitted fallback), and the APCACH inverse-composition approach for guaranteed-contrast color generation. Cite when an agent reaches for "pick a color then check contrast" — the inversion is the load-bearing move.
- ▌ Session Heartbeat · thrillmadeUse when a session must keep working across a stretch longer than one uninterrupted sitting — an orchestrator rolling through a multi-slice plan for hours, a run that will cross a usage-limit reset, a session resumed after a limit or a compaction with someone else's work in flight, or the moment before dispatching an agent when you do not know whether the remaining window can finish it. Also use when a long run is accumulating wakeups but no landed commits, when you are polling for a result the harness would have pushed to you, or when a resumed session is re-deriving pipeline state it should have read off a checkpoint. Names the per-beat order, the largest-dispatch threshold rule, the checkpoint's required slots, and the resume-as-survivor protocol. Not for a single task inside one sitting, and not the unattended-operation policy — that is `unattended-operation`, which builds on this.
- ▌ Technical Writing · thrillmadeFlag a PR that ships a new env var, CLI flag, config key, exported function/type, or endpoint with no matching doc or changelog entry — working and undocumented, not broken. Apply to any addition, a new capability with no removal or behavior change to existing surface. Not for a reversed or superseded value (see retiring-a-superseded-decision) or a breaking API change (see api-contract-enforcement); pairs with the latter when a finding also touches behavior.
- ▌ Udts Linter Rules · thrillmade[L2 stub — incubating] UDTS's machine-enforceable linter rules over its token catalog. Full content is being authored in thrillmade/tokenomics and lands here via an upstream PR once the UDTS spec stabilizes; do not treat this stub as guidance yet.
- ▌ Brand Voice Review · thrillmadeReview user-facing strings for brand-voice consistency. Catch dead phrases ("click here", "you should"), accidental shouting (exclamation marks, ALL CAPS), and jargon that leaks past the design team. Apply to button labels, error messages, marketing copy, headings — not to internal code identifiers or logs.
- ▌ Composing A Screen · thrillmadeEntry point for MAKING a screen — deciding what is primary, what groups with what, what shares an edge, and what waits behind a click, before any pixel, token, or component is picked. Names the composition sequence (rank → group → encode → space → lay out → conventionalise → target → clear the floors → defer) and carries the principles nothing else in this catalog owns: visual hierarchy and the attribute that must separate each level on its own, proximity as a between-to-within gap ratio rather than a pixel count, column count and measure, alignment as an axis count, and progressive disclosure with the three tests anything hidden must pass — plus Jakob's Law (standard / convention / confusion), Fitts's Law (cut the distance before you grow the target), and Norman's two gulfs (gulf of execution: is the next action findable; gulf of evaluation: is its result visible). Use when asked which design principles apply to a screen, when a layout has to be composed from scratch, when ranking actions before picking butt
- ▌ Logmind · thrillmadeMUST be loaded for any task in a project that uses logmind (detect by: .logmind/config.yml at repo root, or AGENTS.md / CLAUDE.md mentioning logmind, or docs/decisions-branches/ present). Use BEFORE writing >20 lines of new code, BEFORE choosing between alternatives, BEFORE adding a dependency, BEFORE modifying existing functionality, BEFORE making any security or performance trade-off, BEFORE renaming or moving any significant module. Logging is part of the work, not after it. Also use to read prior decisions before starting any task in such a project so you don't re-litigate something already decided.
- ▌ Skillforge · thrillmadeSUPERSEDED — skill authoring is now split three ways, and this skill duplicates all three without matching any of them: `skill-creator` (Anthropic plugin) owns measurement, since it carries the only executable eval harness in the set — trigger-rate testing on a held-out split, with-skill vs baseline runs, benchmark aggregation; `superpowers:writing-skills` owns wording form, including the match-the-form-to-the-failure model; and the studio's `skill-smith` agent owns house rules — catalog conformity, the frontmatter contract, upstream-PR discipline. Kept unchanged during the migration window; new work should load the successors. Its own conventions do not fit this catalog: the `.skills/<name>/` layout matches neither `.claude/skills/` nor `skills/`, its `.skills-registry.json` and `skdd forge` tooling are not wired here, and its "under 200 lines" is a third size rule alongside Anthropic's 500 lines and this catalog's CI-enforced byte budget. Use when you need the historical text; otherwise load a successor.
- ▌ Type Scale · thrillmadeUse when designing, auditing, or extending a typography size scale built on a modular ratio. Names the canonical preset ratios (1.067 minor second → 1.618 golden), the picking heuristic by context (UI vs editorial vs marketing), the stops-up / stops-down convention (−2 to +8 default), the rounding rule (round to nearest integer px, rem derived after), and the upper-bound on ratios for product UI. Cite when an agent proposes a ratio above 1.5 for a product system or hand-picks scale steps outside a modular ratio.
- ▌ Dtcg Format · thrillmadeUse when authoring, validating, or transforming W3C Design Tokens Format (DTCG) JSON. Names the $type / $value / $extensions separation, alias-reference syntax ({path.to.token}), group inheritance for shared $type / $description, composite token shapes (shadow, typography, transition), and the namespaced $extensions mechanism for tool-specific metadata. Cite when an agent proposes bare value / type keys without $, invents a top-level $modes key (themes/modes aren't in the DTCG draft), or hand-rolls tool metadata outside a namespaced extension bag. For one system's concrete extension schema see udts-dtcg-extensions.
- ▌ Clud Bug Brand Voice · thrillmadeBrand voice for the clud-bug product surface — naturalist field-guide aesthetic, irreverent without twee, no joke-explainers. Apply to marketing copy, site/, README, docs/. Don't apply to internal logs or code identifiers.
- ▌ Critical Issues Only · thrillmadePR review discipline - flag only correctness, security, and performance issues. Skip nits.
- ▌ Evidence Based Review · thrillmadeEvery PR review claim must quote the specific code being criticized. No hand-waving.
- ▌ Clud Bug Collaboration · thrillmadeHow Claude Code agents working in a clud-bug-installed repo should interact with the bot's review threads, strict-mode gate, and skill set. Use this skill whenever you're about to push a commit, address a clud-bug PR review comment, edit anything under .claude/skills/, modify .github/workflows/clud-bug-*.yml, or wonder why a PR check is red. Also use when planning work in a repo that has a `clud-bug-review` workflow installed — even if the user didn't mention clud-bug by name.
- ▌ Respect Existing Conventions · thrillmadeDon't suggest changes that fight the codebase's established patterns. Match what's already there.