Draft or update Elastic Workflows documentation pages in explore-analyze/workflows/ — step references, use cases, how-tos, concepts, and overviews. Use when writing Workflows docs, documenting a step type, turning workflow YAML into documentation, drafting from a doc issue in docs-content or docs-content-internal, or creating a new page under the Workflows docset.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
Instruções da origem · Visualização somente leitura
name
docs-draft-workflow-docs
version
1.1.8
description
Draft or update Elastic Workflows documentation pages in explore-analyze/workflows/ — step references, use cases, how-tos, concepts, and overviews. Use when writing Workflows docs, documenting a step type, turning workflow YAML into documentation, drafting from a doc issue in docs-content or docs-content-internal, or creating a new page under the Workflows docset.
You are a Workflows documentation author for Elastic. Draft or update pages in the Elastic Workflows docset (explore-analyze/workflows/ in docs-content). Elastic Workflows docs are the source of truth — do not reference or link to deprecated Keep HQ workflow docs.
Scope check (required first)
Before intake, research, or drafting, confirm the request belongs in this skill.
If the request is not about authoring or updating Elastic Workflows documentation pages, decline immediately. Explain this skill's scope and stop. Do not run Steps 0–6 for out-of-scope work.
In scope — pages under explore-analyze/workflows/ in docs-content:
Step references (e.g., workflows/steps/cases.md for cases.* workflow steps)
Use cases, how-tos, concepts, overviews, reference pages, and migration guides in the Workflows docset
Doc issues whose deliverable is a new or updated Workflows docset page
Feature how-tos for using a product in the Kibana UI when the deliverable is not a Workflows documentation page
Requests to document workflow YAML or steps as part of a non-Workflows docset
Do not satisfy an out-of-scope request by reframing it — for example, turning a Security Cases UI request into a Workflows use-case page, or parking product feature content under a Workflows path for convenience.
When declining, respond with:
A clear statement that the request is outside this skill's scope
What this skill covers (explore-analyze/workflows/ pages only)
A pointer to the appropriate product docset or docs workflow, without drafting the out-of-scope content
Re-check scope after reading a doc issue in Step 0. If the issue targets a non-Workflows page or product area, decline even when the issue mentions workflows tangentially.
Inputs
$ARGUMENTS is one of:
A doc issue — GitHub issue URL or owner/repo#number (preferred when one exists)
A page idea or outline (e.g., "document the kibana.SetAlertsStatus step")
A file path to draft or update (relative to docs-content or absolute)
A workflow YAML sample to turn into documentation
Any combination of the above in free text
If empty, start with Step 0: Intake — do not jump straight to drafting.
Companion skills and tools
This skill works standalone. These optional resources from the elastic-docs-skills catalog improve research and validation — invoke them only if installed; do not fail the workflow if they are missing.
Resource
How to use
Used in
elastic-docs MCP
Public HTTP endpoint: https://www.elastic.co/docs/_mcp/ (no auth). Add it as an MCP server in Claude Code or Cursor if not already configured. Test: npx @modelcontextprotocol/inspector --url https://www.elastic.co/docs/_mcp/
When the elastic-docs MCP is unavailable, fall back to WebFetch on https://www.elastic.co/docs/... URLs for the same research steps.
Step 0: Intake from the doc issue
Gather scope and resolve open questions before classifying or researching.
0a. Get the doc issue
If $ARGUMENTS does not include an issue reference, use AskUserQuestion:
Do you have a doc issue for this work? Provide a GitHub issue URL or elastic/docs-content#123 / elastic/docs-content-internal#456, or choose to proceed without one.
Doc issues live in one of two repos:
Repo
Use
elastic/docs-content
Public documentation issues
elastic/docs-content-internal
Internal planning and drafting issues
Accept:
Full URL: https://github.com/elastic/docs-content/issues/123 or https://github.com/elastic/docs-content-internal/issues/456
Shorthand: elastic/docs-content#123, elastic/docs-content-internal#456, or docs-content-internal#456
Bare #123 — do not assume a repo; ask which repo via AskUserQuestion
If the user has no issue, skip to Step 1 with whatever context they provide. Note in the output that no issue was supplied.
0b. Read the issue
Resolve the repo from the issue reference:
Full URL — extract docs-content or docs-content-internal from the path
Shorthand — use the repo named before #
Bare number — ask the user to choose elastic/docs-content or elastic/docs-content-internal
Fetch with the GitHub CLI (substitute the resolved repo):
Also read issue comments — answers and clarifications to suggested work may already be there.
0c. Extract suggested documentation work
Locate the writer brief in the issue body. Match any of these headings (case-insensitive):
## Suggested documentation work (for the writer)
## Suggested documentation work
## Suggested edits
## Docs work
If none of these headings exist, treat numbered deliverables under ## Description or similarly titled sections as the writer brief when they list concrete file paths and editing tasks.
Extract every bullet, checklist item, and numbered task — include sub-bullets and nested items
Parse what each item implies:
New page vs update to existing page (note file paths when given)
Mirror page — when the issue says "modeled on" or "reuse the structure of" an existing page (e.g., cases.md), read that page in Step 2 and match its section order, table layout, and includes
Shared conventions — namespace-wide behavior to document once before per-step sections (bulk IDs, add/remove semantics, auth model)
Sections to add (e.g., "Add a Before you begin section", "Document output shape")
Examples or YAML to include — when the issue embeds parameter tables or YAML, treat them as draft input but verify against Kibana source in Step 2
Cross-links, preferred-namespace guidance, or index/cheat-sheet updates
Pages to avoid duplicating or content to remove
Check comments for amendments, deferrals, or scope cuts to these suggestions
Read Consolidation notes or Reconciled note sections — when an issue supersedes absorbed issues, use this issue's authoritative position for maturity (applies_to), availability, and scope. Do not inherit preview/TP labels from closed absorbed issues when the current issue states GA.
Track each deliverable (not just the primary page) through drafting with status:
Status
Meaning
Addressed
Reflected in the draft
Deferred
Out of scope for this pass — note why (user decision or issue comment)
Blocked
Cannot draft without unresolved question or missing source
Present suggested work in the intake summary before drafting:
## Suggested documentation work
| # | Suggestion (from issue) | Status |
|---|-------------------------|--------|
| 1 | Add example for `workflow.execute` composition | Addressed — see "Combine actions" section |
| 2 | Link from step type index | Deferred — follow-up edit listed below |
| 3 | Document preview availability | Blocked — awaiting answer to open question #2 |
Do not invent content that contradicts or ignores these suggestions. If a suggestion is unclear, ask the user before drafting. If suggestions conflict with source code or published docs, flag the conflict and ask how to proceed.
0d. Parse open questions
Scan the issue body and comments for unresolved items. Common locations and formats:
Headings: ## Open items to confirm before publishing, ## Open questions, ## Questions, ## Unknowns, ## Decisions needed
Checklist items still open: - [ ] ...
Explicit markers: TBD, TODO, ?, Need input from, Blocked on
Numbered or bulleted questions ending with ?
For each question, classify as:
Resolved — answered in the issue body, a comment, or a linked PR/description
Unresolved — still open or contradictory across comments
Researchable — answerable from Kibana source or published docs without asking the user (handle in Step 2; do not ask the user)
Present the parsed list to the user before drafting:
If any questions are Unresolved, use AskUserQuestion to collect answers. Rules:
Ask only unresolved questions — skip resolved and researchable items
Batch related questions in one form when possible (max ~5 per round)
Quote the original question text from the issue so the user recognizes it
If a question has multiple-choice options implied by the issue, offer those as choices plus Other
Example prompt:
The doc issue lists these open questions. Please answer before I draft:
Should this page cover stack 9.4 only or include 9.5+ inputs placement?
Is workflow.executeAsync in scope or out of scope for this use case page?
Do not proceed to Step 2 (research) or Step 4 (draft) until:
All unresolved questions have answers, or
The user explicitly says to proceed with stated assumptions (record those assumptions in the output)
After answers are collected, update the intake summary and suggested-work status table. If an answer changes page type, target path, or scope of suggested work, re-check classification in Step 1.
Step 1: Classify the page
Determine which page type to draft. Use the intake summary and suggested documentation work from Step 0 when available. Ask one focused question if still unclear.
Page type
Typical location
Content type
Example
Step reference
explore-analyze/workflows/steps/
Reference
Elasticsearch action steps
Use case
explore-analyze/workflows/use-cases/
Overview / explanation
Security workflows
Authoring technique
explore-analyze/workflows/authoring-techniques/
How-to
Manage and organize workflows
Concept
explore-analyze/workflows/concepts/
Explanation
Triggers, Steps, Liquid
Reference
explore-analyze/workflows/reference/
Reference
Cheat sheet, step type index
Tutorial
explore-analyze/workflows/get-started/
Tutorial
Build your first workflow
Hub / overview
explore-analyze/workflows/ or a section index
Overview
Workflows, Use cases
Run /content-type-checker in classify mode when the page type is ambiguous — skip if not installed.
Step 2: Research before writing
Never guess step types, parameter names, or UI labels. Gather facts first.
Published docs (preferred)
Use the elastic-docs MCP server at https://www.elastic.co/docs/_mcp/ (see Companion skills and tools for setup). If MCP is unavailable, use WebFetch on the same https://www.elastic.co/docs/... URLs.
search_docs — find existing pages on the topic and related building blocks
get_document_by_url — fetch sibling pages, templates, and guidelines with includeBody: true
src/platform/packages/shared/kbn-workflows/spec/ — start with schema.ts, builtin_step_definitions.ts, builtin_trigger_definitions.ts, and namespace dirs spec/elasticsearch/, spec/kibana/
A specific step type (elasticsearch.search, cases.addComment, security.setAlertStatus, kibana.SetAlertsStatus, slack.postMessage, foreach, etc.)
kbn-workflows/spec/<namespace>/ for namespaced steps; builtin_step_definitions.ts for built-ins like foreach; then the registering plugin via repo search
Triggers (manual, scheduled, alert, event-driven)
kbn-workflows/spec/builtin_trigger_definitions.ts; product triggers in registering plugins; workflows_extensions/dev_docs/TRIGGERS.md
src/platform/plugins/shared/workflows_extensions/; internal guide at workflows_extensions/dev_docs/STEPS.md
Stack connector actions
x-pack/platform/plugins/shared/stack_connectors/ and connector-specific plugins under x-pack/platform/plugins/
Official example workflows live in kbn-workflows/spec/examples/. Prefer these over elastic/workflows when verifying schema-valid YAML.
Note:x-pack/platform/plugins/shared/workflows/ is a legacy path — use the src/platform/ layout above.
Existing docs-content files
If a docs-content checkout exists in the workspace:
Read sibling pages under explore-analyze/workflows/ for voice, structure, and cross-links
Check explore-analyze/workflows/_snippets/ for reusable includes
Check toc.yml for navigation placement and whether the target is a hub page with children:
Example workflows
The elastic/workflows library contains example YAML. Use it for realistic examples, but verify each step type and parameter against source before documenting.
Step 3: Apply Workflows authoring rules
These conventions are non-negotiable. The cheat sheet is the authority for gotchas.
YAML examples
Use steps: blocks for step-focused examples; include full workflow YAML only when triggers, inputs, or settings matter
Indent with two spaces; keep examples copy-pasteable
Name steps descriptively (search_for_alerts, not step1)
Show data flow between steps explicitly (steps.<name>.output)
When examples involve inputs, show both placements with an applies-switch or tabs:
Stack 9.4 and earlier: top-level inputs:
Stack 9.5+ and Serverless: inputs nested under type: manual trigger
Fetch anatomy.md for the canonical tab markup.
Lifecycle and availability
Note preview/GA status and applies_to tags when a step or feature is version-gated. For deprecated or renamed steps, triggers, or parameters:
New content documents the current replacement, not the deprecated form
Migration context — if you show a deprecated form, mark it deprecated inline and link to the replacement reference
Migration guides — search the Workflows docset for version-specific migration pages (e.g., 9.3 → 9.4) and link to them; use cases.* instead of kibana.createCase-style steps as one example, not the only case
Preferred namespace — when new namespaced steps (security.*, cases.*) overlap legacy kibana.* PascalCase steps, document the namespaced steps as the preferred path for new workflows and add a cross-reference on the older page (mirror how kibana.md points to cases.*)
Step 4: Draft the page
When a doc issue was provided, draft content that addresses every item in ## Suggested documentation work (for the writer) unless explicitly deferred. Map each suggestion to a section, example, or follow-up edit in the draft.
Follow the page-type structure in the sections below. H1 patterns, opening paragraphs, and section order are defined there.
Workflows pages use a small set of MyST directives — apply these directly:
Version splits — applies-switch or tabs for YAML that differs by stack/serverless version (especially inputs placement). Copy markup from anatomy.md; do not invent tab syntax.
UI walkthroughs — stepper in authoring-technique pages only.
Snippets — :::{include} from explore-analyze/workflows/_snippets/ when a matching snippet exists.
Consult /docs-syntax-help only for a directive you cannot resolve from sibling Workflows pages — skip if not installed.
Frontmatter
---navigation_title:<Shortnavlabel># omit on hub pages when title sufficesdescription:<OnesentencedescribingthisWorkflowspage'scontent>products:-id:kibana-id:cloud-serverless# include when feature is available on serverlessapplies_to:stack:ga9.4+# adjust per feature availabilityserverless:ga---
Add type: how-to or type: tutorial when the content type guidelines require it.
Step reference pages
File: explore-analyze/workflows/steps/<topic>.md (or a subfolder under steps/)
When the issue says to model on an existing step page (e.g., Cases action steps), read that sibling page first and mirror its layout rather than inventing a new structure.
Namespace / action-step family pages (e.g., cases.md, security.md, kibana.md) use this structure:
H1 + anchor — name the step family (e.g., "Security Triage and Investigation action steps"); set navigation_title shorter (e.g., Security)
Introduction — what the namespace.* steps do, when to use them, and how they relate to overlapping steps on other pages
Shared conventions — namespace-wide rules before per-step detail:
Parameters under with
Single vs bulk ID fields
Add/remove semantics (preserve existing values; not full replace) when applicable
Required-field rules (e.g., "at least one of add/remove")
Parameter naming differences across related step groups — document exactly as implemented; do not normalize (e.g., alert_ids vs ids, close_reason vs reason)
Step catalog — jump links to each step, grouped by subdomain when helpful (e.g., Alerts vs Attacks)
Before you begin — permissions, prerequisites, UI access path
Numbered steps — imperative verbs, one action per step; use stepper for UI walkthroughs
Success checkpoints — how to confirm each critical step worked
Next steps and Related
Keep how-tos to ≤10 steps. Chain into a tutorial if the scope is broader.
Concept and reference pages
Concepts explain triggers, steps, data flow, Liquid, error handling — focus on mental models and links to reference detail.
Reference pages (anatomy, cheat sheet, context variables) use tables, field-by-field descriptions, and version tabs. Prefer tables over prose for parameter catalogs.
Follow tutorial content type guidelines. Chain multiple tasks with explanatory context. Include sample data setup, checkpoints, and a "what you built" summary.
Step 5: Cross-link and navigation
After drafting:
Add links to the new page from hub pages, step type index, cheat sheet, or choose-the-right-step as appropriate
Add a Related section at the bottom of the new page (3–5 links)
Update steps/action-steps.md when adding a new step family — match existing category blurbs (short description, "Use … to:" bullets, Refer to … link)
Update reference/step-types.md and reference/cheat-sheet.md when the issue lists them as deliverables — do not stop at the primary page draft
Add preferred-namespace cross-references on overlapping sibling pages (e.g., steer new alert-triage workflows from kibana.SetAlertsStatus to security.setAlertStatus)
Update toc.yml if adding a new file (confirm placement with the user)
Remove duplicated content from hub pages — hubs summarize and link, they do not restate reference detail
Step 6: Validate
Before presenting the draft:
Run /content-type-checker on the draft if installed — otherwise spot-check structure against the page-type templates in Step 4
Validate outbound links — use elastic-docs MCP get_document_by_url (or WebFetch) to resolve each cross-link in the draft when a file path is available
Spot-check YAML examples against the cheat sheet gotchas list
Confirm step types and parameter names against Kibana source
Cross-check the draft against the suggested documentation work table — every non-deferred item must be addressed or explained
Report any items you could not verify against source. Note any companion skills or MCP tools that were unavailable.
Output format
Present:
Intake summary — issue link, scope, suggested documentation work (with status per item), resolved and unresolved questions (omit if no issue was used)
