Slack Block Kit Skill | Agent Skills

Slack Block Kit

UI framework for building rich, interactive layouts in Slack messages, modals, and App Home.

CRITICAL: Two Markup Systems

Text inside Block Kit text objects uses Slack mrkdwn syntax (*bold*, <url|text>), NOT standard Markdown. The only exception is the markdown block, which accepts standard Markdown (**bold**, [text](url)). Mixing these up is the single most common Block Kit generation error.

Quick Decision Trees

"Should I use blocks?"

Response type?
├─ Conversational reply, short answer, <3 lines   → text only (no blocks)
├─ Multi-section summary, report, dashboard        → blocks
├─ Two-column key-value data                       → blocks (section fields)
├─ Tabular data                                    → blocks (table / data_table)
├─ Charts or metrics visualization                 → blocks (data_visualization)
├─ Code with heading or surrounding context        → blocks
├─ Visual separation needed between topics         → blocks
└─ Feedback buttons or interactive elements        → blocks

"Which block type?"

What am I rendering?
├─ Large section title                → header (plain_text, 150 chars max)
├─ Body text or key-value pairs       → section (text + fields + accessory)
├─ Small metadata or secondary info   → context (images + text, 10 max)
├─ Horizontal separator               → divider
├─ Buttons, menus, date pickers       → actions (25 elements max)
├─ Status callout in a MODAL          → alert (modals only, 200 chars)
├─ Compact entity or summary preview  → card (optional image/actions)
├─ Multiple comparable cards/options  → carousel (1-10 cards)
├─ Grouped/collapsible block set      → container (title + ≤10 child blocks)
├─ Standalone image                   → image (image_url or slack_file)
├─ Formatted text with lists, quotes  → rich_text (nested sub-elements)
├─ Simple static table                → table (100 rows, 20 cols)
├─ Sortable/paginated table           → data_table (caption + 100 data rows)
├─ Pie/bar/area/line chart            → data_visualization (native charts)
├─ LLM-generated markdown content     → markdown (standard MD, messages only)
├─ Embedded video player              → video (requires links.embed:write)
├─ Remote file reference              → file (read-only, source: "remote")
├─ Feedback thumbs up/down            → context_actions (messages only)
├─ Collecting user input              → input (label + element)
├─ AI agent task steps                → plan (sequential tasks, messages only)
└─ Single task with status            → task_card (inside plan or standalone)

"mrkdwn or markdown block?"

Content source?
├─ Short formatted text, labels, fields     → mrkdwn in section/context
├─ Long-form LLM-generated content          → markdown block (standard MD)
├─ LLM-generated tables/task lists/code     → markdown block
├─ Programmatic tabular data                → table or data_table block
├─ Need headings                            → markdown block or header blocks
└─ Mixed: structured layout + prose         → section/header blocks + markdown block

"table, data_table, or markdown table?"

├─ LLM already emitted a Markdown table          → markdown block
├─ Static data, no interaction needed            → table
└─ Users need sorting, paging, or clickable cells → data_table (caption required)

Block Types Overview

Full property tables for all 21 blocks: references/BLOCKS.md. Highlights and sharp edges below.

header

Large bold text. plain_text only — mrkdwn is silently ignored. Max 150 chars.

{ "type": "header", "text": { "type": "plain_text", "text": "Section Title", "emoji": true } }

section

Primary content block. Supports text (3000 chars), two-column fields (10 items, 2000 chars each), and one accessory element. Either text or fields is required. Set expand: true to prevent "see more" truncation on long AI responses.

{
  "type": "section",
  "fields": [
    { "type": "mrkdwn", "text": "*Status:*\nActive" },
    { "type": "mrkdwn", "text": "*Owner:*\nChris" }
  ]
}

Compatible accessories: button, overflow, datepicker, timepicker, select menus, multi-select menus, checkboxes, radio buttons, image.

divider

{ "type": "divider" }

context

Small, muted text for metadata. Elements: mrkdwn text objects or image elements. Max 10 elements.

{ "type": "context", "elements": [{ "type": "mrkdwn", "text": "Last updated: Feb 9, 2026 • deploy-bot" }] }

actions

Interactive elements: buttons, select menus, overflow menus, date pickers. Max 25 elements.

{
  "type": "actions",
  "elements": [
    { "type": "button", "text": { "type": "plain_text", "text": "Approve" }, "style": "primary", "action_id": "approve_action", "value": "approved" }
  ]
}

Button styles: primary (green), danger (red), or omit for default. Use primary sparingly — one per set. Action IDs must be unique within the message.

alert

Callout for status, risk, or urgency. Modals only — putting an alert in a message payload is invalid. Text accepts plain_text or mrkdwn, max 200 chars. level: default, info, warning, error, or success. For message-surface callouts, use a section block with an emoji or a card instead.

{
  "type": "alert",
  "text": { "type": "mrkdwn", "text": "*Dependency conflict detected* before deploy." },
  "level": "warning"
}

card

Compact, scannable preview for entities, summaries, records, or agent results. At least one of hero_image, title, actions, or body is required.

{
  "type": "card",
  "title": { "type": "mrkdwn", "text": "Daily Standup Reminder" },
  "subtitle": { "type": "mrkdwn", "text": "Runs every weekday at *9:00 AM*" },
  "body": { "type": "mrkdwn", "text": "Last run: Today at 9:00 AM. Status: Success" },
  "actions": [
    { "type": "button", "text": { "type": "plain_text", "text": "View Logs" }, "action_id": "view_logs" }
  ]
}

Fields: icon/hero_image (image elements), slack_icon (built-in icon, mutually exclusive with icon), title/subtitle (150 chars), body/subtext (200 chars), actions (max 3 buttons; danger left-aligns, primary/unstyled right-align). No size attribute.

carousel

Horizontal, scrollable group of card blocks for options, recommendations, search results, or next steps. Must contain 1-10 cards.

{
  "type": "carousel",
  "elements": [
    { "type": "card", "title": { "type": "mrkdwn", "text": "Option A" } },
    { "type": "card", "title": { "type": "mrkdwn", "text": "Option B" } }
  ]
}

container

Groups up to 10 child blocks under a required plain_text title (150 chars). Optional subtitle, icon, width (narrow/standard/wide/full), and collapsing (is_collapsible, default_collapsed). Child blocks: actions, context, divider, file, header, image, input, rich_text, section, table, video — no nested containers, cards, or plan blocks.

{
  "type": "container",
  "title": { "type": "plain_text", "text": "Bulk update: 2 records" },
  "is_collapsible": true,
  "child_blocks": [
    { "type": "section", "text": { "type": "mrkdwn", "text": "*DCW-1024*\nStatus: Open → Closed" } }
  ]
}

image

Standalone image with required alt text. Provide either image_url (public, max 3000 chars) or slack_file object. Formats: png, jpg, jpeg, gif.

{ "type": "image", "image_url": "https://example.com/chart.png", "alt_text": "Deployment success rate chart" }

rich_text

Advanced formatted text with nested elements: styled text, lists, code blocks, quotes. See references/RICH-TEXT.md before writing nested lists or inline mentions — the nesting rules are easy to get wrong.

{
  "type": "rich_text",
  "elements": [
    { "type": "rich_text_section", "elements": [{ "type": "text", "text": "Key findings:", "style": { "bold": true } }] },
    {
      "type": "rich_text_list",
      "style": "bullet",
      "elements": [
        { "type": "rich_text_section", "elements": [{ "type": "text", "text": "Latency reduced by 40%" }] }
      ]
    }
  ]
}

Sub-element types: rich_text_section (paragraph), rich_text_list (style: "bullet" or "ordered"), rich_text_preformatted (code block), rich_text_quote (blockquote). Inline types: text, link, emoji, user, channel, usergroup, broadcast, date, color.

table

Simple tabular data. Each row is a flat array of cell objects (NOT an object with a cells property). There is no columns property — the first row is the header. Cell types: raw_text, raw_number, rich_text. Max 100 rows, 20 columns, 10,000 chars across all cells (per table AND aggregate per message).

{
  "type": "table",
  "rows": [
    [{ "type": "raw_text", "text": "Service" }, { "type": "raw_text", "text": "Status" }],
    [{ "type": "raw_text", "text": "API" }, { "type": "raw_text", "text": "Healthy" }]
  ],
  "column_settings": [{ "align": "left" }, { "align": "center" }]
}

data_table

Interactive table: pagination, sorting, filtering, clickable cells. Same row shape as table, plus a required caption string. Rows: header + 1–100 data rows; all rows need equal cell counts; header cells cannot be rich_text. Optional page_size (1–100, default 5) and row_header_column_index (default 0). Numeric sorting when a column is all raw_number cells.

{
  "type": "data_table",
  "caption": "Service health",
  "rows": [
    [{ "type": "raw_text", "text": "Service" }, { "type": "raw_text", "text": "Latency (ms)" }],
    [{ "type": "raw_text", "text": "API" }, { "type": "raw_number", "value": 12, "text": "12" }]
  ]
}

data_visualization

Native pie, bar, area, or line charts. Required: title (≤50 chars) and chart. Pie: segments (1–12, labels ≤20 chars, values > 0). Bar/area/line: series (1–12; names ≤20 chars, unique) plus axis_config.categories — every data point label must exactly match a category, and no series may omit a category.

{
  "type": "data_visualization",
  "title": "Weekly Sales",
  "chart": {
    "type": "bar",
    "series": [
      { "name": "Online", "data": [{ "label": "Week 1", "value": 32000 }, { "label": "Week 2", "value": 41000 }] }
    ],
    "axis_config": { "categories": ["Week 1", "Week 2"], "y_label": "USD" }
  }
}

markdown

Standard Markdown rendering for AI app output. Messages only.

{ "type": "markdown", "text": "**Bold**, *italic*, [link](https://example.com)\n\n## Heading\n\n- List item" }

Supports bold, italic, strikethrough, links, headers (all levels render the same size), lists, code (inline and fenced with syntax highlighting), block quotes, dividers, tables, task lists, and escaping. Images render as hyperlink text. Cumulative 12,000-char limit across all markdown blocks per payload. block_id is ignored. One markdown block may translate into multiple Slack blocks.

context_actions

Feedback and icon buttons for message-level actions. Messages only. Max 5 elements. Compatible elements: feedback_buttons, icon_button (only icon: trash).

video

Embedded video player. Requires links.embed:write scope and a publicly accessible URL in the app's unfurl domains.

input

Collects user data in modals, messages, and Home tabs. Requires label (plain_text, 2000 chars) and one compatible element. See references/ELEMENTS.md for all input element types.

plan

Container for sequential task cards, designed for AI agent output. Messages only.

{
  "type": "plan",
  "title": "Thinking completed",
  "tasks": [
    { "task_id": "t1", "title": "Fetched data", "status": "complete" },
    { "task_id": "t2", "title": "Generating report", "status": "in_progress" }
  ]
}

Task status values: pending, in_progress, complete, error. Each task is a task_card block with optional details, output (rich_text), and sources (url elements).

file

Remote file reference. Read-only — cannot be directly added to messages by apps.

Streaming Agent Output

Use chat.startStream, chat.appendStream, and chat.stopStream for live AI responses (all require chat:write). Start requires channel and thread_ts — streamed messages should reply to a user request. When streaming to channels, recipient_user_id and recipient_team_id are also required.

chunks (accepted by all three methods) can include:

markdown_text chunks — standard Markdown text
task_update chunks — timeline-style task progress (same shape as task_card; id not task_id)
plan_update chunks — update a plan title
blocks chunks — arrays of Block Kit blocks (max 50 per array; extras are dropped with a warning)

task_update and plan_update chunk fields are limited to 256 characters.

Set task_display_mode on chat.startStream:

timeline (default): tasks appear individually in sequence
plan: tasks grouped in one plan; first task placement determines plan placement
dense: consecutive tool calls collapse into a single summarized task card

chat.stopStream can add final blocks rendered after streamed content. Its 50-block limit is separate from streamed blocks chunks, allowing up to 100 total finalized blocks. Rate limits: start/stop Tier 2 (20+/min), append Tier 4 (100+/min).

Composition Objects

Option object (used in selects, overflow, checkboxes, radio buttons):

{
  "text": { "type": "plain_text", "text": "Option 1" },
  "value": "opt_1",
  "description": { "type": "plain_text", "text": "Detailed description" }
}

Text max 75 chars, value max 150 chars, description max 75 chars.

Confirmation dialog (add to any interactive element via confirm):

{
  "title": { "type": "plain_text", "text": "Are you sure?" },
  "text": { "type": "plain_text", "text": "This action cannot be undone." },
  "confirm": { "type": "plain_text", "text": "Yes, do it" },
  "deny": { "type": "plain_text", "text": "Cancel" },
  "style": "danger"
}

Others: conversation filter (include: im/mpim/private/public), dispatch action config (on_enter_pressed, on_character_entered), Slack file, Slack icon (card blocks), trigger/workflow (workflow buttons). Full property tables: references/COMPOSITION.md.

Limits

| Constraint | Limit | |-----------|-------| | Blocks per message | 50 | | Blocks per modal/Home tab | 100 | | Section text | 3000 chars | | Section fields | 10 items, 2000 chars each | | Header text | 150 chars | | Context elements | 10 | | Actions elements | 25 | | Context actions elements | 5 | | Alert text (modals only) | 200 chars | | Card title/subtitle | 150 chars | | Card body/subtext | 200 chars | | Card action buttons | 3 | | Carousel cards | 1-10 | | Container child blocks | 10 | | Container title/subtitle | 150 chars | | Table rows / columns | 100 / 20 | | Table chars (per table and per message) | 10,000 | | Data table rows | header + 100 data rows | | Data viz series/segments | 1-12 (labels/names 20 chars) | | Data points per series | 1-20 | | Markdown block text | 12,000 chars cumulative per payload | | Streaming chunk fields (task/plan update) | 256 chars | | Modal title / submit / close | 24 chars each | | Modal views in stack | 3 | | Modal private_metadata | 3000 chars | | Button text | 75 chars (displays ~30) | | Button value | 2000 chars | | action_id / block_id | 255 chars | | Overflow options | 5 | | Select options | 100 | | Option text | 75 chars | | Placeholder text | 150 chars | | File input max file size | 10MB per file |

Anti-Patterns

| Anti-Pattern | Problem | Fix | |--------------|---------|-----| | Blocks without text fallback | Empty notifications, no accessibility fallback | Always provide text in chat.postMessage | | text and blocks diverge | Notification says one thing, chat shows another | Keep semantically aligned | | Blocks for simple replies | Visual noise for short responses | Use text only for simple replies | | alert block in a message | Alerts are modal-only; payload rejected | Use section + emoji or card in messages | | Huge tables in one message | 10,000-char cell limit exceeded | Split across multiple messages | | Rows as { "cells": [...] } objects | Invalid — rows are flat arrays of cells | Each row = array of cell objects | | columns property on table | No such property | First row of rows is the header | | Chart labels not matching categories | Runtime validation failure | Every data point label must match axis_config.categories | | mrkdwn in header text | Ignored — headers only accept plain_text | Use plain_text type | | Standard Markdown in mrkdwn fields | **bold** and [links](url) render literally | Use *bold* and <url\|text>, or a markdown block | | Long header text | Silently truncated at 150 chars | Keep under 150 | | Missing alt_text on images | Accessibility failure, API may reject | Always include alt_text |

Best Practices

Use blocks when: the response has multiple distinct sections; two-column key-value layouts improve readability; a table or chart presents data more clearly than prose; visual separation helps; code needs surrounding context; interactive elements are needed.

Don't use blocks when: the response is conversational ("sure, done"), under ~3 lines, or a simple answer to a direct question.

Always:

Provide a complete text field as the accessible fallback (notifications, threads, search, screen readers)
Keep the text and blocks semantically aligned
Use mrkdwn syntax in text objects, not standard Markdown (except in markdown blocks)
Escape &, <, > in user-generated content

Surfaces Overview

| Surface | Max Blocks | Key Methods | Notes | |---------|-----------|-------------|-------| | Messages | 50 | chat.postMessage, chat.update | Primary output surface | | Modals | 100 | views.open, views.update, views.push | Requires trigger_id (3s expiry), up to 3 stacked views | | App Home | 100 | views.publish | Private per-user view, Home/Messages/About tabs | | Canvases | N/A | canvases.create, canvases.edit | Markdown only — no Block Kit support | | Lists | N/A | lists.* API methods | Task tracking and project management | | Split View | N/A | Agents & AI Apps config | AI chat surface with Chat + History tabs |

Modals collect input via input blocks and return view_submission payloads. private_metadata (3000 chars) persists context between views. Read references/SURFACES.md before building modal flows or App Home views — it covers the view object schema, response actions, and view.state.values extraction paths.

Work Objects

Work Objects render rich entity previews when links are shared in Slack. They extend link unfurling with structured data, flexpane details, editable fields, and actions.

| Type | Entity ID | Purpose | |------|-----------|---------| | File | slack#/entities/file | Documents, spreadsheets, images | | Task | slack#/entities/task | Tickets, to-dos, work items | | Incident | slack#/entities/incident | Service interruptions, outages | | Content Item | slack#/entities/content_item | Articles, pages, wiki entries | | Item | slack#/entities/item | General-purpose entity |

Work Objects use chat.unfurl with a metadata parameter containing entity type, external reference, and entity payload. Read references/WORK-OBJECTS.md before implementing — Slack silently drops the whole payload if any required field is missing.

Reference Documentation

| Read this | Before doing this | |-----------|-------------------| | references/CHEATSHEET.md | Quick lookup of any block/element/limit at a glance | | references/BLOCKS.md | Building any block payload — all 21 blocks with full property tables | | references/ELEMENTS.md | Adding buttons, selects, inputs, or pickers — all 20 elements | | references/COMPOSITION.md | Writing text objects, options, confirm dialogs, filters, Slack icons | | references/RICH-TEXT.md | Writing rich_text blocks — nesting, inline types, styles | | references/SURFACES.md | Building modals, App Home, canvases, or split view flows | | references/WORK-OBJECTS.md | Implementing link unfurls, flexpanes, editable fields |

Sources

Block Kit Reference — Slack
Block Kit Blocks — Slack
Block Kit Elements — Slack
Composition Objects — Slack
chat.startStream — Slack
chat.appendStream — Slack
chat.stopStream — Slack
Work Objects — Slack
Surfaces — Slack
Modal views — Slack
App Home — Slack

Agent Skills: Slack Block Kit

Install this agent skill to your local

Skill Files