Flow Tables logo

Flow Tables

Community
jpoindexter
flow-tables

Use when designing or reviewing any data table, list, or feed — choosing table vs list vs cards, adding sorting/filtering/bulk actions/pagination, or handling empty, loading, and error states. Also applies when users report tables being unusable on mobile, losing their place on refresh, or when a screen must display more than ~10 structured records.

Overview

Publisherjpoindexter
Repositoryux-flow-skills
Skill nameflow-tables
Stars
5
Forks
0
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by jpoindexter on GitHub. Read the source before you install it.

Installation

Install the Flow Tables AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/jpoindexter/ux-flow-skills/tree/main/skills/flow-tables
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/jpoindexter/ux-flow-skills.git /tmp/ux-flow-skills
mkdir -p .claude/skills
cp -r /tmp/ux-flow-skills/skills/flow-tables .claude/skills/flow-tables
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Flow Tables in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Flow Tables on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Flow Tables is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Data Tables, Lists & Feeds

Overview

A table is a workspace, not a printout: users scan, sort, select, and act in place. The proven pattern is dense rows + progressive disclosure of actions + persistent user control — sort, filters, and column config survive refresh.

When to Use

  • Any screen displaying >10 structured records users must scan, compare, or act on
  • Adding sort, filter, column config, bulk select, or inline editing to an existing table
  • Choosing pagination vs infinite scroll vs load-more
  • Designing empty, loading, and error states for data-heavy views
  • Adapting a wide table for mobile
  • Reviewing a table users describe as "cluttered," "slow," or "I lost my place"
  • NOT for ≤5 items with no comparison need — use a simple list or cards
  • NOT for search-results ranking and filter chips (see flow-search)

The Proven Flow

  1. Choose the display form first.
    • Table — many attributes per record, cross-column comparison matters, dense scanning. Airtable, Stripe payments, admin panels.
    • List — one primary attribute + metadata, vertical scan, per-row triage. Gmail threads, Linear issues, notification centers.
    • Cards — the visual IS the content (images, products, boards). Never cards for records users compare field-by-field — cards put fields in different positions and kill comparison.
    • Feed — reverse-chronological, append-only, consumed rather than managed. Activity logs, social timelines.
  2. Toolbar above the table.
    • Search/filter input left → applied-filter chips → density/column/view controls right.
    • One row, sticky together with the header. Proven by Linear, Airtable, Stripe Dashboard.
  3. Sorting.
    • Click a header to sort; click again to reverse; visible arrow on the active column only (arrows on every column hide the active one).
    • Default sort must be meaningful — newest first for activity, name for directories — never insertion order.
    • Multi-column sort is a power feature behind a menu, not the default interaction.
    • Persist the user's sort choice per view.
  4. Filtering.
    • Applied filters render as removable chips above the table ("Status: Open ×"), plus "Clear all" once ≥2 are active.
    • Filters combine AND across fields, OR within a field. Result count updates live ("128 results").
    • Persist filters in the URL (?status=open&assignee=me) so views are shareable and refresh-safe — Linear and GitHub Issues prove this.
    • Offer saved views once users repeat the same filter combos (Linear views, Gmail labels-as-views).
    • Power tools graduate to a filter builder (field + operator + value rows) — Airtable, Linear — but chips remain the visible summary.
  5. Row actions.
    • Primary action = clicking the row (opens detail or peek panel).
    • Secondary actions appear on hover at row-end: icon buttons, max 3, plus an overflow ⋯ menu. On touch, the ⋯ is always visible.
    • Hover reveals must not reflow the row — reserve the space.
    • Destructive actions live only inside the overflow menu, confirmed or undoable — never bare on the row.
  6. Bulk select.
    • Checkbox column at far left (40–48px wide); row checkbox appears on hover (desktop) or always (touch).
    • Header checkbox selects the visible page, with an explicit "Select all 1,204 matching" link — Gmail's pattern; never silently select the full result set.
    • On selection, a floating action bar appears (bottom-center, or replacing the toolbar): "12 selected — Archive · Assign · Delete · ×".
    • Shift-click selects ranges; Esc clears the selection; the selected count is always visible while any selection exists.
  7. Inline editing.
    • Click-to-edit a cell (Airtable): the cell becomes an input in place; Enter/blur commits, Esc reverts, Tab moves to the next editable cell.
    • Save optimistically; on failure, roll back the cell and show a toast with retry.
    • Reserve inline editing for flat values (text, number, select, date). Anything multi-field opens the detail panel instead.
    • Precedence: inline-edit tables demote row-open to the identity-column link or an explicit open affordance — never both whole-row-click and cell-edit on the same table.
  8. Row → detail.
    • Peek panel (Linear) or push navigation. Selection is reflected in the URL; Back closes the peek.
    • Keyboard ↑↓ or j/k moves selection; Enter opens; the focused row stays scrolled into view.
    • Make rows real links so cmd/ctrl-click opens detail in a new tab.

Pagination vs infinite scroll vs load-more

  • Pagination — users need to return to a spot, share a location, reach the end, or act on ranges. Admin tables, transactions, anything with bulk actions. Show 1 … n, prev/next, and a page-size picker (25/50/100). Stripe Dashboard's choice.
  • Infinite scroll — leisure consumption feeds only (social, discovery). Forbidden when there's a footer, bulk actions, or a "find it again" need. Must restore scroll position on back-navigation.
  • Load-more button — the safe middle: user-controlled, footer stays reachable, works for search results and mobile lists. Default to this when unsure.
  • Use cursor-based loading under the hood for live data — offset pagination skips or duplicates rows when the data changes between pages.
  • Live-updating feeds don't shove new rows in while the user reads — show a "3 new items" pill at the top that inserts on tap (Twitter/X; Slack's "new messages" divider is the sibling pattern).

Empty, loading, error (all three are mandatory)

  • First-use empty: icon/illustration + one line of value + primary CTA — "No invoices yet — Create your first invoice". Never a blank region or a bare "No data".
  • Filtered-to-empty: a different message — "No results match these filters" + one-click "Clear filters". Never show the first-use CTA here; the user has data, just not under these filters.
  • Loading, first load: 5–8 skeleton rows matching real row height and column layout — mismatched skeletons cause layout jump on arrival.
  • Loading, refetch: keep stale data visible with a subtle progress indicator. Never blank a table the user is reading.
  • Error: inline in the table region — "Couldn't load payments" + Retry button, filters preserved. Never render an empty table as if there were genuinely no data when the fetch failed.
  • Partial failure (some rows/columns failed to hydrate): render what loaded, mark the failed cells, offer retry — don't fail the whole view.

Feeds specifically

  • Feed item anatomy: avatar/icon left; actor + action + object as the first line; timestamp right or below; optional preview block.
  • Group repeated events ("Ana and 4 others commented") — ungrouped event spam drowns the signal.
  • Relative timestamps ("2h ago") with absolute on hover/tap; switch to absolute dates past ~7 days.
  • Unread semantics must be explicit and singular: bold + dot for unread, cleared on view or on click — pick one rule and keep it.

Layout & Structure

  • Row heights: compact 32–36px, default 40–44px, comfortable 48–56px. Offer a density toggle in pro tools (Gmail does). Touch rows ≥44pt.
  • Header row: sticky on scroll; 40–44px tall; sortable columns show their affordance on hover; header labels never wrap to 3 lines — shorten the label.
  • Alignment: text left; numbers/amounts right-aligned with tabular figures; dates left in one consistent format.
  • Status: colored badge + text label — never color alone (accessibility), never more than ~6 distinct status colors per table.
  • Column widths: first (identity) column widest and sticky during horizontal scroll; fixed widths for dates/status/amounts; long text truncates with ellipsis + tooltip.
  • Columns user-resizable in data-heavy tools, persisted per user per view.
  • Column config: gear/⋯ in the toolbar → show/hide + drag-reorder; persist per user per view. Ship with ≤7 visible default columns.
  • Counts to display: total near the toolbar ("1,204 payments"), selected count in the action bar, filtered count after filtering.
  • Row grouping (by status, assignee, date): sticky group headers with counts, collapsible; Linear's default view proves it beats one undifferentiated list.
  • Summary/footer row for financial tables: totals pinned at the bottom, right-aligned under their columns.
  • Zebra striping: skip it at default density with adequate row padding; use hover highlight instead. Borders: horizontal separators only; vertical rules only in spreadsheet-like grids.
  • Mobile adaptation (<768px): collapse each row to a card — identity field as title, 2–3 explicitly chosen priority fields as labeled lines, status badge top-right, ⋯ for actions.
  • Never horizontal-scroll a 9-column table on a phone; choose the priority columns deliberately.
  • Alternative for genuinely tabular data on mobile: sticky identity column + horizontal scroll with a visible affordance (edge fade or scrollbar).
  • Performance: virtualize lists >100 rows; paginate or window anything >1k rows; debounce filter inputs 150–300ms.
  • Numeric formatting: thousands separators, consistent decimals per column, unit in the header not in every cell ("Amount (USD)").
  • Links in cells: one primary link per row (the identity column); other cells plain text unless genuinely navigable.
  • Timestamps: one format per table — relative for activity tables, absolute for records and audit trails.
  • Checkbox column: 40–48px fixed width; its count mirrors into the bulk-action bar.

Quick Reference

SituationPattern
Many attributes, field comparisonTable
One key attribute + metadata triageList rows
Visual-first contentCard grid
Users must find a row again / share positionPagination
Leisure consumption feedInfinite scroll
Unsure / mobile / search resultsLoad-more button
New items arriving while user reads"N new items" pill, insert on tap
Actions on many rowsCheckbox column + floating bulk-action bar
Selecting beyond the visible pageExplicit "select all N matching" link
Editing one flat valueInline cell edit, optimistic save
Editing multiple fieldsOpen the detail panel
Repeated filter combosSaved views
Long undifferentiated listGroup rows by status/date with sticky headers
Table on a phoneCard collapse with 2–3 priority fields
Fetch failedInline error + Retry (never fake-empty)
Filters applied, zero rows"No matches" + Clear filters
Brand-new account, zero recordsValue line + create-first CTA
Activity / audit trailFeed with grouped events + relative timestamps
Money/quantity columnsRight-aligned, unit in header, tabular figures
Unread tracking in a listBold + dot, one clear rule for clearing

Anti-Patterns

Anti-patternWhy it breaksFix
Actions column with 5 always-visible icon buttonsVisual noise; destructive actions one mis-click awayRow click = primary; hover reveals ≤3; rest in ⋯; delete only in ⋯
Infinite scroll + bulk actions or footerFooter unreachable; selection scope unboundedPagination or load-more
Blanking the table during refetchUser loses reading position and contextKeep stale rows; overlay a subtle loading indicator
"No data" shown on fetch errorUser believes records are gone; support tickets followDistinct error state with Retry
Filters/sort reset on refresh or backUsers rebuild their view constantlyPersist view state in the URL
Header checkbox silently selecting all 10k rowsAccidental mass actions on unseen dataSelect visible page + explicit "select all matching"
New rows shoving in mid-readUser loses their place; mis-clicks follow"N new items" pill, insert on tap
Horizontal-scrolling 9 columns on mobileOne column visible at a time; unusableCard collapse with priority fields
Centered or left-aligned numeric columnsDigits don't align; magnitude comparison failsRight-align with tabular figures
Unsorted default orderTable appears random; trust dropsMeaningful default sort with visible arrow
Cards for record comparisonFields sit in different positions per cardUse a table
Skeletons unlike real contentLayout jumps on load (CLS)Skeleton matches final row height and columns
Row actions appearing only on hover, on touchTouch has no hover; actions unreachableAlways-visible ⋯ on touch
Hover actions that reflow the rowContent shifts under the cursor; mis-clicksReserve the action-area space
Confirm dialog for every single-row actionTrains users to click through confirmationsConfirm only destructive/irreversible; prefer undo toast
Color-only status indicatorsFails colorblind users and grayscale scansBadge = color + text label
Every cell rendered as a blue linkRow reads as link soup; primary action unclearOne primary link per row; rest plain text
Mixed timestamp formats in one tableUsers can't compare times across rowsOne format per table, applied consistently
Ungrouped event spam in feeds40 rows of "X edited Y"; signal drownsGroup repeated events by actor/object

Frequently asked questions

What does the Flow Tables AI skill do?

Use when designing or reviewing any data table, list, or feed — choosing table vs list vs cards, adding sorting/filtering/bulk actions/pagination, or handling empty, loading, and error states. Also applies when users report tables being unusable on mobile, losing their place on refresh, or when a screen must display more than ~10 structured records.

Why use Flow Tables on TypingMind?

Because you install it once and use it with any model. Flow Tables is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Flow Tables in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/jpoindexter/ux-flow-skills/tree/main/skills/flow-tables. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Flow Tables?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Flow Tables?

As many as you like. As long as a model supports skills, you can use Flow Tables with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Flow Tables AI skill free?

Yes. It is published on GitHub by jpoindexter under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇