write-docs-integration-guide

name	write-docs-integration-guide
description	Write an integration guide for the Tiger Data Docs site following established conventions. Use when documenting how to connect a third-party tool (database client, BI tool, streaming platform, observability stack, ETL connector, etc.) to Tiger Cloud or self-hosted TimescaleDB.
user-invocable	true
allowed-tools	Read Grep Glob Edit Write Bash Agent
argument-hint	[product name or source URL]
effort	high

Write a Tiger Data Docs integration guide

End state: a .mdx file at src/content/docs/integrate/<category>/<slug>.mdx, a sidebar entry in astro.config.ts, and (optionally) a logo asset. Most guides are bilateral: configuration on the third-party side, configuration on the database side, and optionally a step that triggers data flow. Short integrations may collapse to one part; long ones may have three or four.

In every AskUserQuestion call, treat options as equal — no (Recommended) labels.

Start by reading two files

Before writing anything, read:

1. src/content/docs/integrate/_template-integration.mdx — the canonical skeleton you copy from. This file is the source of truth for placeholder wording in scaffold mode.
2. One existing page in the same category as a tone/depth reference:
   • BI/visualization → bi-vizualization/tableau.mdx
   • Data engineering/ETL → data-engineering-etl/apache-kafka.mdx (long, multipart) or debezium.mdx (short, two-part)
   • Observability/alerting → observability-alerting/grafana.mdx
   • Workflow orchestration → data-engineering-etl/apache-airflow.mdx

Match the depth and tone of an existing page in the same category. Don't invent a new structure.

Identify the product

Pin down the canonical product name and its kebab-case slug before anything else — they reappear in title, intro, prereqs, verify section, success line, sidebar label, file path, and logo filename.

• If a product name was passed as argument, normalize it (kafka → Apache Kafka, slug apache-kafka) and confirm via AskUserQuestion.
• Otherwise ask: "What's the canonical product name?" and derive the slug (lowercase, spaces and / → -, punctuation dropped).

Record:

• Product name — vendor's preferred casing (Apache Kafka, Power BI, pgAdmin). Used in prose, title, sidebar label, success line.
• Product slug — kebab-case. Used in file path, logo filename, RelatedContentCard href.
• Vendor URL — for the one-liner.

Pre-fill exactly one sentence at the top of the body: a vendor-linked one-liner about what the product is. Example: "Apache Kafka is a distributed event streaming platform used for high-performance data pipelines, streaming analytics, and data integration." Stop there. The "This page shows you how to..." sentence is procedural content — handle it per mode below.

Modes of operation

Ask via AskUserQuestion which mode applies:

Mode	Choose when
Scaffold	The user knows the integration steps and wants the template pre-processed — frontmatter resolved, structure laid down, placeholders to fill in.
Full draft	The agent writes the body content from a source URL (vendor docs, GitHub example, blog post) or general product knowledge.

The deployment-scope question, cloud-platform follow-up, logo question, sidebar entry, and pnpm build pass run in both modes.

Scaffold mode

Resolve from up-front answers: frontmatter (title, description, integrationPlatforms, sidebar label, category), imports, prereq partial choice, product name everywhere it appears, vendor one-liner, success line, sidebar entry.

For everything procedural — part headings, NumberedItem titles, body lines, code blocks, verify steps — copy _template-integration.mdx verbatim, preserving every parenthesized placeholder for the user to fill in. Don't invent part headings ("Configure your database" is a guess unless the user said so), and don't write SQL, CLI commands, or connection examples.

If you find yourself typing a real heading like "Create a hypertable" or a real CREATE TABLE, you've drifted into full-draft mode — stop.

Full-draft mode

After the up-front questions, ask the user for a source: vendor integration page URL, GitHub example repo, existing draft, or blog post.

• Source provided: adapt it. Extract real commands, sequence, prereqs, then write a full draft. In your final reply, list parts you couldn't verify against the source.
• No source: improvise from general product knowledge. In your final reply, state explicitly that the content is unverified and every command needs to be reviewed.

Warnings about uncertainty go in your reply to the user, not into the published page as Callout blocks. The published doc reads as confident prose with placeholders only where the user still needs to fill in.

Confirm deployment scope

Ask via AskUserQuestion which Tiger Data deployment(s) this integration supports. The answer drives frontmatter, prereq partial, title, prose constants, and whether to use <Tabs>.

Scope	integrationPlatforms	Prereq partial	Title pattern	Constants in prose	<Tabs>?
Tiger Cloud only	cloud platforms (follow-up)	_prereqs-cloud-no-connection.mdx	Integrate <Product> with Tiger Cloud	{C.SERVICE_LONG}, {C.CLOUD_LONG}	No
Self-hosted only	[self-hosted]	_prereqs-self-instance.mdx	Integrate <Product> with TimescaleDB	{C.SELF_LONG}, {C.TIMESCALE_DB}	No
Cloud-primary, self-hosted parity assumed	cloud platforms + self-hosted	_prereqs-cloud-and-self.mdx	Integrate <Product> with Tiger Cloud	Lead with {C.SERVICE_LONG}; partial adds parity note	No
Both equally supported, identical steps	cloud platforms + self-hosted	_prereqs-cloud-or-self.mdx	Integrate <Product> with Tiger Data	"your service or database"	No
Both equally supported, diverging steps	cloud platforms + self-hosted	_prereqs-cloud-or-self.mdx	Integrate <Product> with Tiger Data	Per-tab: {C.CLOUD_LONG} / {C.SELF_LONG}	Yes; labels Tiger Cloud / Self-hosted TimescaleDB

If the scope is "Both equally supported", follow up with AskUserQuestion to distinguish identical steps (single body, no <Tabs>) from diverging steps (page-level <Tabs>, see Bilateral parts).

If the scope includes Tiger Cloud, follow up to pin cloud platforms: AWS only → [aws], Azure only → [azure], both → [aws, azure]. Skip when the scope is Self-hosted only.

File location

src/content/docs/integrate/<category>/<slug>.mdx. <category> is one of:

Folder	For products that...
bi-vizualization	Visualize data or build dashboards (Tableau, Power BI)
code	Help developers write app code against the database
configuration-deployment	Provision, deploy, or manage infrastructure
connectors	Are Tiger Data native source/destination connectors
data-engineering-etl	Move, transform, or orchestrate data (Kafka, Airflow, Debezium)
data-ingestion-streaming	Ingest or stream data into the database (Fivetran, HiveMQ)
observability-alerting	Monitor, alert on, or scrape metrics/logs (Grafana, Prometheus, Datadog)
query-administration	Manage queries, schemas, or admin tasks
secure-connectivity	Add network or auth security on top of connections

If no category fits, ask before inventing a new folder.

Frontmatter

Match existing pages — strings unquoted unless they contain : or [, arrays inline:

---
title: Integrate <Product name> with <Tiger Data product>
description: One-sentence SEO-optimized description, sentence case, ends in a period
integrationCategory: <category folder slug>
integrationPlatforms: [<subset of: aws, azure, self-hosted>]
integrationIndustry: [<industry 1>, <industry 2>, <industry 3>]
keywords: [<topic 1>, <topic 2>, <topic 3>, <topic 4>, <topic 5>]
integrationCardLogo: <product-slug>.png
sidebar:
  label:   

• integrationPlatforms is a strict enum (aws, azure, self-hosted) validated by src/content.config.ts.
• integrationIndustry is free-form — 2–3 realistic verticals, don't pad.
• keywords — 4–7 phrases readers would actually search for; include the product name plus its function.
• integrationCardLogo is optional. See Logos and images.

Imports — required vs optional

Always required:

import * as C from "@constants";
import { Prerequisites } from "@components/Prerequisites";
import { NumberedList, NumberedItem } from "@components/NumberedList";
import IntegrationPrereqs from "@partials/<path-to-prereq-partial>.mdx";

Pick the prereq partial based on deployment scope:

Partial	Use when
_prereqs-cloud-and-self.mdx	Same steps work on Tiger Cloud and self-hosted
_prereqs-cloud-or-self.mdx	Reader picks Tiger Cloud OR self-hosted (two equal paths)
_prereqs-self-instance.mdx	Self-hosted only
_prereqs-cloud-no-connection.mdx	Console-only workflow inside Tiger Cloud

Also import <ConnectionDetails /> when the procedure asks the reader to plug host/port/dbname/user/password into the third-party tool. The decision is about what the steps below require, not which prereq partial you picked.

Only import below what you actually use — the build flags unused imports:

import { Callout } from "@stainless-api/docs/components";                                  // scope callouts, warnings
import { RelatedContentCards, RelatedContentCard } from "@components/RelatedContentCards"; // Next steps section
import ConnectionDetails from "@partials/_prereqs-connection-details.mdx";                 // inside Prerequisites
import { Image } from "astro:assets";                                                      // optimized images
import { Tabs, TabItem } from "@astrojs/starlight/components";                             // parallel paths (Cloud vs self-hosted)

Page layout (top → bottom)

1. Intro paragraph — what the product does, plus one sentence on what this page shows.
2. Optional scope Callout (variant="note") — only when the product has multiple integration paths and this page covers one of them. Example: "This page covers the self-managed Apache Kafka path. If your Kafka cluster runs on Confluent Cloud, use the fully managed connector in {C.CONSOLE} instead."
3. "In this integration guide, you:" bullets — short verb-led outcomes.
4. <Prerequisites context="integration"> — <IntegrationPrereqs />, then <ConnectionDetails /> if applicable, then product-specific bullets (account, install, version). Always link installation pages.
5. One or more parts — see Bilateral parts.
6. ## Verify the integration — two concrete end-to-end steps; one usually triggers something on the third-party side, the other confirms on the database side.
7. Success line — "You have successfully integrated with ."
8. Optional ## Troubleshooting — specific gotchas, one-line cause/fix; link /integrate/troubleshooting for general issues.
9. Optional ## Limitations — what isn't supported.
10. Optional ## Next steps — RelatedContentCards linking to related pages.

Bilateral parts

Most integrations split into two parts: one for the third-party side, one for the database side. Use ## headings that name the outcome — never ## Step 1:. The NumberedList inside the section does the numbering.

Typical shapes:

• Configure <tool> + Prepare your service — Kafka (install + sink connector on tool side, hypertable on db side)
• Configure your database + Configure <tool> — Debezium (publication on db side, connector on tool side)
• Single bilateral step — Tableau (one part is enough)

If the integration has a discrete data-movement step (send a test message, run a DAG, kick off a sync), that becomes a third part.

Inside each part:

<NumberedList>
  <NumberedItem title="Verb-phrase title for the sub-step">

    What the reader does. Use [connection details](/integrate/find-connection-details) when referring to `host`, `port`, `dbname`, `user`, `password`.

    ```sql
    -- code if needed
    ```

    What the reader should see or verify after this sub-step.

  </NumberedItem>
</NumberedList>

If a sub-step has its own numbered sequence, write every item as 1. inside the NumberedItem body — the renderer styles them as a., b., c. at the nested level. Don't nest a second NumberedList.

Diverging Tiger Cloud vs self-hosted steps

When the scope is "Both equally supported, diverging steps", wrap the entire procedural body — from the first part heading through ## Verify the integration — in a single page-level <Tabs> block. Shared content (intro, outcomes, <Prerequisites>, Troubleshooting, Limitations, Next steps) stays outside.

<Tabs>
  <TabItem label="Tiger Cloud">
    {/* parts + verify section, copying placeholder wording from _template-integration.mdx; use {C.CLOUD_LONG} */}
  </TabItem>
  <TabItem label="Self-hosted TimescaleDB">
    {/* same shape; use {C.SELF_LONG} */}
  </TabItem>
</Tabs>

Tab labels are the Tiger Data product names exactly: Tiger Cloud and Self-hosted TimescaleDB. Reuse the template's placeholder wording inside each <TabItem> — don't invent new phrasing for the tabbed variant.

Constants — required

import * as C from "@constants";

Use the matching constant for every product, brand, or term it covers. The full list lives in src/constants.ts — read it directly. Common ones: {C.PG} (PostgreSQL), {C.CLOUD_LONG} (Tiger Cloud), {C.SERVICE_LONG} (Tiger Cloud service), {C.SELF_LONG} (self-hosted TimescaleDB), {C.TIMESCALE_DB}, {C.CONSOLE}, {C.HYPERTABLE}, {C.COMPANY}.

Never hardcode a product or brand name in prose when a constant exists. The TigerData.ProductConstants Vale rule (run by the vale.yml CI check) nudges literal product/brand names toward their constant, including {C.PG} for PostgreSQL/Postgres. Inside backticks or URLs, the literal string is fine.

In NumberedItem titles, use a template literal when interpolating:

<NumberedItem title={`Connect to your ${C.SERVICE_LONG}`}>

Preserve inline backticks in titles — they render correctly.

Authoring rules

• Headings are sentence case. Proper nouns and acronyms excepted. The Google.Headings Vale rule (run by the vale.yml CI check) enforces this.
• No ## Step N: prefix. NumberedList handles numbering; use plain verb-phrase headings.
• No --- horizontal rules. Section headings already separate content.
• No Latinisms. "for example" not "e.g.", "that is" not "i.e.", "and so on" not "etc.", "versus" not "vs.".
• No emojis unless asked.
• Every code block needs context. Tell the reader where to run it, what it does, what to expect.
• Cross-link liberally. Link /integrate/find-connection-details when referring to host/port/etc.; link concept pages in /learn/... when introducing concepts.
• No backwards-compat scaffolding. Delete optional sections that don't apply.
• No deprecated compression APIs. See Use current columnstore/hypercore APIs.

Use current columnstore/hypercore APIs

TimescaleDB 2.18.0 renamed the compression API to columnstore/hypercore (compress_chunk → convert_to_columnstore, *_compression_policy → *_columnstore_policy, timescaledb.compress → timescaledb.enable_columnstore, etc.). The old names still work as backwards-compat aliases but must not appear in new integration guides — in prose, SQL, CLI examples, connector configs, dbt models, or screenshots. If you're adapting from a vendor doc or source URL that uses the old names, translate as you write.

Full mapping table, carve-outs (compress_chunk_time_interval and compress_sparse_index are NOT deprecated), and verification grep one-liner live in .claude/references/deprecated-compression-apis.md. Read that file before writing SQL or pasting third-party config, and again before saving the page.

Logos and images

Card logo

Ask via AskUserQuestion whether the user has a logo:

Option	What to do
Light + dark variants	Save both to src/assets/images/integrate/card-logos/ as <product-slug>.png and <product-slug>-dark.png. Set both integrationCardLogo and integrationCardLogoDark in frontmatter.
Light only	Save <product-slug>.png. Set integrationCardLogo in frontmatter; omit integrationCardLogoDark.
No file yet	Omit integrationCardLogo entirely — the overview card falls back to initials in a colored badge. Optionally set integrationCardInitials to override the default initials (for example, λ for AWS Lambda).

Screenshots and illustrations

Save to src/assets/images/integrate/<category>/<descriptive-name>.png. Import with astro:assets:

import { Image } from "astro:assets";
import imgFoo from "../../../../assets/images/integrate/<category>/<file>.png";

<Image src={imgFoo} alt="Descriptive alt text" />

See src/assets/images/README.md for the project's image conventions, including the integration-screenshot layout and light/dark-mode versions.

Components reference

Component	Source / docs
Callout	src/components/Callout.astro — variants are exactly tip, note, important, warning, callout (CTA with button). No info/success/danger/caution — any other value is an authoring error.
NumberedList / NumberedItem	src/components/NumberedList.tsx — NumberedItem accepts title: string
Prerequisites	src/components/Prerequisites.tsx — pass context="integration"
RelatedContentCards / RelatedContentCard	src/components/RelatedContentCard.tsx; per-page right-rail guide in src/components/LearnMore.README.md
Tabs / TabItem	@astrojs/starlight/components (NOT Stainless) — https://starlight.astro.build/components/tabs/
Image	astro:assets — https://docs.astro.build/en/guides/images/

After writing the page

1. Add a sidebar entry in astro.config.ts under the matching category in the "Integrate" tab. Insert alphabetically within the category:

   { label: "<Product name>", link: "/integrate/<category>/<slug>" },
   

2. Place the card logo in src/assets/images/integrate/card-logos/ if the user provided one.

3. Run the build:

   pnpm build
   

   Fix anything flagged — MDX errors show as Element type is invalid: undefined or Expected component X to be defined. Prose, product-constant, and heading checks run separately via Vale (pnpm lint:prose, or the vale.yml CI check on changed lines), not as part of pnpm build.

Checklist before finishing

• File at src/content/docs/integrate/<category>/<slug>.mdx
• Frontmatter has title, description, integrationCategory, integrationPlatforms (subset of enum), integrationIndustry, keywords, integrationCardLogo, sidebar.label
• Required imports present; optional imports only present if used
• Prereq partial matches the deployment scope
• <ConnectionDetails /> rendered if and only if the procedure asks the reader to enter db credentials into the third-party tool
• Intro is one paragraph + "In this integration guide, you:" bullets
• Parts use plain verb-phrase headings — no "Step N:" prefix
• Verify the integration section has two concrete steps
• All product names use {C.CONSTANT} — no hardcoded "PostgreSQL" / "Tiger Cloud" / "TimescaleDB" in prose
• No deprecated compression APIs — grep is clean (see the mapping)
• Headings are sentence case; no --- rules, Latinisms, or emojis
• Every code block has surrounding context
• Card logo handled (file in card-logos/ and integrationCardLogo set, or field omitted for initials fallback)
• Sidebar entry added in astro.config.ts, alphabetical within category
• pnpm build passes