agent-skills

High-retention insight Reels for Instagram (Silph Scope / SilphCo Analytics). Create or optimize faceless, AI-voice Pokémon TCG market insight Instagram Reels — price/volume analysis, myth-busting set narratives, concentration stats, collector mental models. Optimizes strictly for 2026 Instagram Reels ranking signals (skip rate / first 1–3s, watch time, DM shares, saves). Apply when writing Instagram Reel scripts, reviewing or ranking finished .mp4 Reels for algorithm impact, improving retention, designing share triggers, or generating new Silph Scope Instagram episodes. Routed from /instagram. When ranking which Instagram Reel will travel farthest, this skill overrides data-reels craft density. Use when the user runs /high-retention-insight-reels-instagram, /instagram, or mentions Instagram Reels, Silph Scope, or insight reels.

Instagram routing hub for SilphCo / Silph Scope. Use when the user runs /instagram or asks about Instagram Reels, Reels ranking, skip rate, shares, captions, Insights diagnosis, or which Instagram short to ship. Owns the native metrics map (skip/share/like/save/repost/comment) and post-publish primary-action decision tree. Routes to high-retention-insight-reels-instagram for faceless TCG insight Reels create/review/rank. Prefer this hub for any Instagram-first task so specialists stay discoverable under one slash command.

Principles for high-retention short-form data video (Instagram Reels, TikTok, Shorts) built from measured competitor research on data-viz and stats accounts. Use when making or reviewing data-driven short-form video, reels, TikToks, voiceover-timed data animations, retention/format questions for social video, or when a project's own reel playbook needs a philosophy layer to sit on top of. This is concepts, not commands — pair with your project's own reel playbook (operational scripts, render pipeline, asset paths) for execution.

Routing hub for marketing — offers, StoryBrand messaging, direct-response ads, word-of-mouth, social virality, and Silph Scope insight Reels. Use when creating or reviewing landing pages, pricing, emails, ads, lead magnets, launches, referral loops, Instagram Reels, or social content, or when the right marketing skill is unclear. Routes to marketing-offers, marketing-storybrand, marketing-cashvertising, marketing-contagious, marketing-going-viral, high-retention-insight-reels-instagram. Prefer product-design for product app UI craft; quality-check for ship/e2e; data for pipelines. Prefer this hub over loading all marketing specialists at once.

Production Flutter UX audit: ranked polish findings (dead taps, null text, a11y, keyboard, haptics, layout shift, async/offline honesty, motion, empty states, destructive confirm, permission priming, RTL, charts/number consistency) with optional PLAN.md per pick. Use for Flutter app polish, "feels off", release UX pass, /flutter-pro-ux-review, /flutter. Complements official Flutter architecture/test skills — does not replace them. Not for greenfield layering, RenderFlex-only fixes, React/web craft, or CI setup.

Routing hub for Flutter app work — structure vs polish vs tests vs layout fixes. Use when the task is Flutter/Dart mobile UI and the right specialist is unclear, or user says /flutter. Routes to official flutter-* architecture, test, layout skills when present, and flutter-pro-ux-review for production UX polish. Not for React/web product craft (product-design) or pure backend.

Routing hub for product UI/UX craft — density, hierarchy, mobile chrome, design-system fidelity, interaction quality, explore-before-build, and product chart judgment. Use when polishing app surfaces (card, chat, trade, filters, HUDs), sticky/mobile UX, empty/loading/error states, "works but feels off", motion/micro-interaction review, or when the right craft skill is unclear. Routes to product-ui-craft, mobile-product-ux, design-system, web-quality, ui-explore, mockup-implement, tufte, emil-design-eng, make-interfaces-feel-better, 12-principles-of-animation, flutter-pro-ux-review (Flutter apps), then browser-verify. Prefer frontend-design for full FE implementation pipelines (perf, SPA structure) when the ask is build-the-feature not craft. Do not use for marketing landings (marketing), pure QA/e2e (quality-check), or data pipelines (data).

Mobile-readable **product ECharts** (card price/volume, portfolio history, set charts) without desktop regressions — grid gutters, axisLabel floors, legend placement, adaptive height, touch tooltips, dual-viewport proof. Use when card or portfolio charts are clipped, tiny, sparse, legend-jammed, "mobile echarts", or /mobile-app-echarts-visual. Prefer mobile-blog-chart-visual for blog SVG. Prefer design-system chart tokens + tufte for honesty.

Mobile-readable **blog / article SVG (and static) charts** without wrecking desktop density — viewBox/fontSize scale formula, soft type floors, label wrap, plot-share, fixed right value column, no max-h / overflow-hidden clip, no chart subtitles under the title. Use when blog figures are squinty, clipped, sparse, "mobile chart", "fix blog charts", or /mobile-blog-chart-visual. Prefer mobile-app-echarts-visual for product card/portfolio ECharts. Prefer tufte for chart-type honesty; browser-verify for e2e proof.

Alias for mobile-blog-chart-visual. Use /mobile-blog-chart-visual for blog SVG charts; /mobile-app-echarts-visual for product ECharts.

Write original educational and long-form technical content in a specific author's authentic voice — courses, lessons, explainers, deep-dives, threads — using a two-layer voice pack (an always-on PRIMARY voice that reproduces the author's idiolect + SECONDARY craft borrowed from master writers, applied not impersonated, routed by the piece's job) and a facts-first workflow that verifies technical content BEFORE styling. In this agent-skills install the default pack is evan (Evan Kim; profiles/evan/). The kaue pack (Superteam Brazil) remains available. Use whenever drafting or editing content that should sound like a specific person rather than generic AI — 'write this in my voice', 'in Evan's voice', 'in Kaue's voice', or 'draft a thread the way <author> would'. Also builds new voice packs via the persona-builder agent. Prefer the writing hub when the writing path is unclear; use writing-prose for house human tone without a named persona; writing-technical for form without full persona; writing-docs for pure p

Routing hub for writing that sounds human and stays coherent — Evan default voice pack, house anti-slop prose, technical articles, pure docs, and marketing handoff. Use when drafting or editing lessons, posts, threads, essays, docs, research notes, or anything that must not read like generic AI; when the user wants Evan's voice or a specific persona; or when the right writing skill is unclear. Routes to writer-style (default pack: evan), writing-prose, writing-technical, writing-docs, marketing. Prefer product-design for product UI craft; quality-check for ship/e2e.

Technical writing form — multi-mode structures for findings, field logs, claim diaries, bake-offs, design briefs, concept maps, and forum notes. Mechanism-first, evidence objects, reframe insights, open gaps. Domain-agnostic (not only data/math). Prefer writing-docs for pure procedures; writing-prose for general human tone; writer-style + evan pack for full Evan cadence and mode routing; writing hub when unclear.

Pure documentation and reference writing — runbooks, README install paths, API/tool reference, how-to procedures, operator notes. Scannable structure, prerequisites, failure modes, copy-paste commands. Use for docs that must be followed, not essays. Prefer writing-technical for research-style explainers; writing-prose for opinionated posts; writer-style for named voice; marketing for landing copy. Prefer writing hub when the writing path is unclear.

House writing floor — human naturalness, anti-slop, coherent opinionated structure without persona cosplay or LLM word-soup. Use when drafting or editing posts, docs, essays, threads, READMEs, or technical explainers that should sound like a sharp human, not generic AI, and no named voice pack is required. Prefer writer-style (default pack: evan) for a named voice. Prefer writing-technical for research/build-log form; writing-docs for pure procedures. Prefer marketing for offer/story/ad frameworks. Prefer writing hub when the writing path is unclear.

Prove UI in a real browser — Playwright e2e, visual snapshots, axe, and/or Chrome DevTools MCP (live DOM, console, network, performance). Use only when the task is proof, regression, debugging runtime paint, or "claim done" needs pixels. Prefer quality-check when the QA path is unclear. Prefer frontend-design when still building UI. Not for writing product styles from scratch (design-system / product-ui-craft). Aliases: visual-verify, browser-testing-with-devtools.

Routing hub for implementing product UI in code — SPA structure, React performance, mockup ports, and the full build pipeline. Use when building or shipping product UI features, card/chat/trade surfaces, code-splitting, or when the right FE engineering skill is unclear. Prefer product-design for pure craft (density, mobile chrome, “feels off”, tokens/a11y polish without a full build). Prefer quality-check for prove/e2e/ship. Prefer this hub over frontend-ui-engineering. Do not use for marketing copy (marketing) or data pipelines (data).

Port a signed-off mockup (HTML picker winner, Figma, screenshot) into production with fidelity — no freestyle redesign. Use only when a design is already chosen and the task is "match this." Prefer product-design when craft/explore path is unclear; frontend-design for full FE feature implementation. Not for open-ended exploration (ui-explore first) or inventing new visual systems (design-system / craft).

Throwaway design exploration before production — standalone HTML A/B pickers, in-app multi-variant UI prototypes, or tiny terminal logic prototypes. Use when the user wants design options, compare A vs B, "prototype this", "try a few designs", or /html-design. Prefer product-design when multi-step product UX craft is unclear; frontend-design for full FE builds. Not for production ports (mockup-implement after pick), charts (tufte), marketing copy, or ship proof (quality-check / browser-verify). Aliases: html-design, prototype.

Audit and fix interaction quality — a11y, focus, forms, touch targets, semantics, motion preferences. Use only when the task is an a11y/guidelines pass or form/dialog interaction fix. Prefer quality-check when the goal is ship/verify/regression; prefer product-design when multi-step product UX craft is unclear; frontend-design for full FE builds. Not for visual brand redesign, React performance alone (react-performance), or Playwright/e2e proof loops (browser-verify).

Stay on project design-system rails — tokens, typography, components, chart contracts, and brand locks. Use only when the task is already about tokens, DESIGN.md / design-system docs, or agents inventing fonts/colors/spacing. Prefer over generic aesthetic skills. Prefer product-design when multi-step product UX/craft is unclear; frontend-design for full FE implementation. Not for greenfield brand invention, marketing storytelling, general layout polish (product-ui-craft), or pure QA/verify (quality-check).

Mobile product UX — sticky chrome, bottom bars, sheets, safe areas, gesture conflicts, one-thumb density. Use only when phone/tablet chrome is the focus (sticky misbehaving, sheet vs keyboard, adapting desktop product UI to mobile). Prefer product-design when multi-step product UX is unclear (it routes here). Prefer frontend-design for full FE feature builds. Not for marketing landings only, pure desktop dashboards, general polish (product-ui-craft), or browser proof loops (browser-verify / quality-check).

Raise product UI craft — spacing, hierarchy, density, radii, states, and restrained motion so chrome feels designed. Use only when polishing already structured UI ("works but feels off"), not as the default for any UI task. Prefer product-design when multi-step product UX/craft is unclear (it routes here). Prefer frontend-design for full FE feature implementation pipelines. Not for brand invention, marketing landings, design-system token definition (design-system), mobile sticky/sheet systems (mobile-product-ux), or ship/e2e proof (quality-check / browser-verify).

QA routing hub for verification — TDD, diagnose, browser/visual proof, web-quality, doubt-driven review, check-work, PR review, ship checklists, data-semantic-quality. Use when the QA path is unclear, when shipping or fixing consumer-facing regressions (search, filters, sticky chrome, URL/ debounce), or on "QA", "verify", "prove it", "e2e", "flaky", "don't ship bugs". Prefer specialists directly when already clear (only tdd, only browser-verify). Prefer frontend-design when the work is still building/redesigning product UI (hand off here for proof). Do not use for pure product design exploration, marketing, issue filing alone (qa), or lakehouse design without verification intent (data).

Alias for ui-explore (HTML A/B + prototype explore). Prefer loading ui-explore. Use when user says html-design, HTML mockups, A/B design picker. Prefer frontend-design when multi-step product UI is unclear.

Alias for browser-verify mode B (Chrome DevTools MCP). Prefer loading browser-verify. Use when user says DevTools MCP, live DOM, console, network. Prefer quality-check when QA path is unclear. Not for general UI build (frontend-design).

Alias for ui-explore (throwaway logic/UI prototypes). Prefer loading ui-explore. Use when user says prototype, try a few designs, state machine sandbox. Prefer frontend-design when multi-step product UI is unclear.

React/SPA performance — request waterfalls, bundles, re-renders, chart/list jank, code-splitting heavy libs. Use only when load time, jank, or a performance budget is the stated problem. Prefer frontend-design when the task is general UI build/polish (it routes here if needed). Not for pure CSS aesthetics, a11y-only reviews (web-quality), or chart cognitive design (tufte).

Alias for browser-verify (Playwright + DevTools proof). Prefer loading browser-verify. Use when user or older docs say visual-verify, Playwright snapshots, or axe e2e proof. Prefer quality-check when QA path is unclear.

Use when designing, modifying, or debugging Apache Iceberg lakehouses (bronze/silver/gold medallion specifically on Iceberg) — PyIceberg + Polars/DuckDB/Arrow/pandas writes, Spark or Flink ingestion, catalog choice (Glue/REST/Polaris/Lakekeeper/Nessie/JDBC), Spark/Trino maintenance procedures, gold aggregates, compaction/expire/orphan, branching/WAP, CDC, snapshot rollback, slow incremental jobs, lock contention, schema evolution. Includes Iceberg internals and PyIceberg metadata-table diagnostics. Don't use for Delta Lake or Databricks-native medallion, Apache Hudi, Snowflake-native, or BigQuery-native designs — even if "medallion" is mentioned. Don't use for plain Parquet on S3 with a Hive metastore (that's not Iceberg), OLTP modeling, generic Airflow/dbt orchestration unrelated to Iceberg, or Postgres/MySQL schema work — different invariants. Prefer the data hub when the right data skill is unclear or the task spans ingest→store→serve.

Use when consuming external HTTP APIs for data ingestion (rate limiting, exponential backoff, pagination/cursors, auth, response schema validation, caching, idempotent landing) or when serving gold/analytical data over HTTP (FastAPI + DuckDB, pushing filters down to the engine, keyset pagination, cache invalidation on publish, factory routers over many gold tables). Also use when defining serving contracts that honor persisted quality attributes, freshness SLIs, publish-token invalidation, or publish-coupled serving sidecars (derived projections rebuilt on source snapshot — not a second quality system). Covers blockchain-indexer / marketplace / price-feed ingestion and lakehouse gold-serving APIs. Don't use for defining semantic quality rules or thresholds (that's data-semantic-quality), generic REST/GraphQL interface design unrelated to data movement (api-and-interface-design), wrapping an API as a live MCP tool, or OLTP/CRUD application backends. Prefer the data hub when the right data skill is unclear or t

Use when building or debugging pipelines that use DuckDB as the embedded analytical engine — tuning memory_limit and threads, handling larger-than-memory queries and disk spilling, reading Parquet/CSV efficiently (predicate & projection pushdown, glob/Hive partitioning, union_by_name), writing Parquet well (PER_THREAD_OUTPUT, ROW_GROUP_SIZE, the partitioned-write temp-blowup trap), connection reuse, and EXPLAIN ANALYZE profiling. Covers DuckDB-over-Parquet and DuckDB-as-query-layer on a single node. Don't use as an OLTP/transactional application database, for cluster-scale work that genuinely exceeds one big node, or for generic SQL tutoring. For querying Iceberg tables (`iceberg_scan`, catalog reads) see data-apache-lakehouse; for serving DuckDB results over HTTP see data-api. Prefer the data hub when the right data skill is unclear or the task spans ingest→store→serve.

Use when running multiple data pipelines/services on shared single-host infrastructure and reasoning about memory admission, concurrency caps, or capacity — sizing systemd MemoryMax/cgroup limits, building an admission gate that queues instead of skipping work, diagnosing OOM kills that only appear when several individually-fine pipelines coexist, or right-sizing caps from measured evidence instead of folklore. Covers subprocess/systemd-run scope accounting, wait-budget-vs-unit-timeout races, and the capacity ratchet loop (observe → cap generously → tighten on evidence). Don't use for single-pipeline internal memory tuning — that's the engine-specific skill (data-duckdb for DuckDB, data-apache-lakehouse's single-host section for PyIceberg writes) — or for Kubernetes/cluster resource management, which has different primitives than one host running several systemd-managed pipelines. Prefer the data hub when the right data skill is unclear or the task spans ingest→store→serve.

Portable methodology for semantic (row-truth) data quality in pipelines — write-time quality attributes, single-sourced scoring, entity-scoped rule evaluation, provenance trust ladders, cohort-relative fences, golden entity packs with dual error budgets, and layered enforcement. Use when designing or debugging quality flags, outlier or anomaly rules, entity-resolution confidence gates, classification trust ladders, producer–consumer quality contracts, split-brain between stored flags and API/UI filters, or golden pack regression for correctness. Don't use for schema/type/null-fraction fences alone (data hub / data-apache-lakehouse), multi-pipeline memory admission or OOM (data-pipeline-operations), table retirement (data-table-lifecycle), DuckDB engine tuning (data-duckdb), or domain-specific business thresholds and product rule books (keep those in the product repo's own skills — not this pack). Prefer the data hub when the right data skill is unclear or the task spans ingest→store→serve.

Use when deciding whether a table or artifact should keep existing, planning a table drop or deprecation, auditing a lakehouse for dead/zero-consumer tables, or building maintenance-job coverage across many tables. Covers the consumers-or-deprecate discipline (every table names a reader or gets flagged for removal), the drop-durability trap (a get_or_create-style helper silently resurrecting a "dropped" table on its next write), the metadata-vs-physical split of a drop (catalog drop_table does not delete files), and generating maintenance coverage from the catalog instead of a hand-maintained list. Don't use for choosing a storage format or catalog backend, or schema evolution within a live table — that's data-apache-lakehouse. This skill is about whether a table should exist at all, and how to retire it safely once it shouldn't. Prefer the data hub when the right data skill is unclear or the task spans ingest→store→serve.

Use when writing, reviewing, or improving online ads, email promotions, opt-in pages, website conversion sections, CTAs, headlines, pricing frames, credibility proof, social proof, cart recovery, or buyer-psychology tactics using Drew Eric Whitman's Cashvertising Online. Best for direct-response conversion improvements after the offer and message are clear. Prefer the marketing hub when the marketing path is unclear; use this skill only for its specific framework.

Use when designing or reviewing word-of-mouth, referral, sharing, PR, campaign, product-launch, social-sharing, or behavior-spread strategy using Jonah Berger's STEPPS framework from Contagious. Helps make products, ideas, messages, and campaigns more likely to be talked about, shared, remembered, and passed along. Do not use as a substitute for customer-message clarity; use marketing-storybrand first when the offer itself is unclear. Prefer the marketing hub when the marketing path is unclear; use this skill only for its specific framework.

Use when designing, researching, scripting, reviewing, or diagnosing social media content for virality using Brendan Kane's Viral Content Model from The Guide to Going Viral, plus Alex Hormozi's hook guidance from $100M Playbook: Hooks. Helps choose platform-native formats, run Gold/Silver/Bronze analysis, identify performance drivers, generate and rank content ideas, improve hooks, retention, storytelling, and conversion. Best for social content strategy, not general brand positioning. Prefer the marketing hub when the marketing path is unclear; use this skill only for its specific framework.

Use when designing, reviewing, or improving offers, packages, pricing, bonuses, guarantees, scarcity, urgency, naming, value proposition, or conversion economics using Alex Hormozi's $100M Offers. Helps make an offer more valuable before writing ads or landing-page copy. Do not use for general brand messaging, viral social formats, or word-of-mouth strategy unless offer architecture is the blocker. Prefer the marketing hub when the marketing path is unclear; use this skill only for its specific framework.

Use when clarifying brand, product, landing page, email, ad, sales, or website messaging with a StoryBrand-style customer narrative. Helps turn unclear copy into simple customer-centered messaging, one-liners, calls to action, lead magnets, drip campaigns, testimonials, and internal mission narratives. Do not use for generic literary story analysis, fiction writing, brand voice work that does not involve customer conversion or organizational alignment, or word-of-mouth/referral mechanics (that's marketing-contagious). Prefer the marketing hub when the marketing path is unclear; use this skill only for its specific framework.