Workflows

Use Workflows to turn repeated tasks into reliable operations. This page helps you create the workflow, run it, debug it, and schedule it.

Questions this page should answer

1. Which recurring task should become a workflow first?
2. Are runs completing with useful outputs, not just green status?
3. Where should we fix steps, inputs, or schedule to improve reliability?

Organize workflows in folders

Use Create folder in the Workflows library. Set a clear name and optional description, then select Save folder. Open a folder card to see its items; use the breadcrumb to return to the library. Use an item’s Move to folder action to choose a destination, or select Top level to return it to the library root. Check the Folder field when creating an item if the form offers it. Moving an item changes its location, not its contents or permissions. Folder changes require write access.

Before you automate

• Start with one repeated task your team already does manually.
• Define the exact output you expect before building.
• Keep version one short and clear.

What this page gives you

• A workflow library with reusable cards.
• Fast creation from templates or from scratch.
• A full editor for step logic and sequencing.
• Per-workflow run, execution, and history views.
• Schedule controls for recurring runs.

How to read the workflow library

In the top section:

• Each card is one automation.
• The card title and subtitle tell you the business job.
• Last or Never run tells you adoption and recency.
• The three-dot action on cards helps with quick maintenance actions.

Use the table below cards as your global operations feed:

• Status: run health.
• AI credits: cost per run.
• Inputs: confirms run context.
• Started and Duration: speed and timing consistency.

When a run fails, inspect Inputs first, then step output.

Create workflow (template or scratch)

Use Create workflow to start from:

• Start from scratch for custom logic.
• Templates for common execution patterns.

Template-first is faster for most teams. Start from scratch only when your flow is unique.

Edit workflow (builder)

The editor is where you design and refine execution:

• Left panel: available step types.
• Center canvas: step order and flow.
• Right panel: AI assistant and properties.
• Top actions: run test, save, and chat-guided edits.

Keep steps explicit. Each step should do one clear job.

Inputs, outputs, and variables

An input is information supplied when the workflow runs, such as a URL or topic. A step output is the result produced by one action. A variable is a named value that another step can reuse. For example, {{generateStructuredPageContent.generated_page}} refers to the generated_page field from that step’s result. Use a fixed value when every run should use the same setting. Map a variable when it should come from the current run or a previous step. Keep the expected type: a full JSON object, an array of items, and a text field are different values. Inspect a test run’s actual output before mapping it into a publishing step. Saving the workflow stores its definition. Running it starts execution. A scheduled run starts execution later, while a scheduled publishing step can set a different publication time. Check each step’s result and external destination before treating a completed workflow as proof that content is live.

Steps tab

Use Steps for individual actions such as prompting an LLM, branching, human review, API calls, content operations, and publishing. Drag a step onto the canvas, then configure its Properties. Review destination and input mappings before running a step that changes external content.

Sequences tab

Use Sequences for reusable research, content creation, enrichment, outreach, and Programmatic Pages operations. Open the matching group and inspect the sequence’s inputs and outputs before connecting it to the rest of the workflow.

Content image enrichment workflows

Use content enrichment steps when a draft already exists and needs useful inline visuals:

• Enrich content images: search the project’s Product Images library first, then use AI web search when no first-party image strongly matches. It adds the strongest images near the sections they support and saves the result as a new active editor version. It skips webpage screenshots.
• Create content version: save markdown from another step as a new active editor version without overwriting older versions.

Product Images keep their existing Atomic storage URL and do not need a public source-credit line. External images are imported into Atomic storage and include a visible credit to the original source. This works well after Generate article or when refreshing older text-only drafts. Add clear titles, descriptions, tags, and use cases on Product Images so the agent can find the right first-party visual. Keep target image count balanced; weak visual opportunities should be skipped instead of forcing filler images.

Repurpose content for distribution

Use Repurpose content after Get content from editor to adapt one finished article into one selected format:

• LinkedIn post or article
• X post or thread
• Newsletter issue

The step applies the current project Brand kit and can use explicitly selected Knowledge Bases for supporting project facts. No Knowledge Base selection means no project Knowledge search. It preserves source claims, numbers, quotes, attribution, and links instead of researching and rewriting the topic from scratch. Repurpose content only returns a structured draft. It does not publish, schedule, authenticate, or create a draft on an external platform. Place Human review after it to approve or edit the draft. The Repurpose content for distribution template already uses this safe generation-and-review sequence and ends with the approved draft as workflow output. For X, choose Post for one complete idea or Thread for ordered segments. If a faithful point cannot fit within the configured X post limit, the step skips the post instead of shortening it into a misleading teaser. For a newsletter issue, choose a publication profile and provide the issue number. Previous-issue details, referral campaigns, commercial links, and other changing values are included only when you supply them. The The Startup Finance profile keeps its navigation, issue shell, rundown, TL;DR, numbered sections, mid-issue CTA position, bottom line, reply CTA, referral module, signature, and P.S. in one reviewable document. The screenshot shows an illustrative copy-only draft from the local example project. Choose No media for a copy-only issue, Use supplied Atomic assets for reviewed project media, or Discover and insert media to let the step reuse project images, generate source-backed charts, and find images or GIFs. Discovered external media is imported into project-scoped Atomic storage before it is embedded. Every inserted asset has a reviewable Atomic URL, alt text, placement, source, and rights status. Reviewers can remove or replace an asset without regenerating the copy. Run details show the exact base and target skill versions used for the draft. They also provide Markdown and HTML exports rendered from the same structured issue, so the visible document and exports cannot drift.

Publish approved social content

Connect the destination under Project settings → Integrations, then choose the publishing step and supported account/channel. The current integration catalog includes Buffer, Narrareach, and Reddit in addition to website publishing. Different steps accept different destinations and formats; use the fields displayed for the selected step. Keep Human review before public publishing. Review the final body or ordered thread segments, destination, and any media. Generating a repurposed draft does not publish it. An immediate publishing step creates an external post; inspect run details and created links before retrying a partial failure.

Schedule Buffer publication

Add Publish via Buffer, select the Buffer connection and Social account, and confirm the displayed network. Enter the approved text and choose the media type supported by that destination. Publication date and time (UTC) controls delivery through Buffer. The workflow sends the post to Buffer when the step executes, with that future publication time. Leave the field blank for immediate publication. Use a UTC date and time, or map a UTC ISO timestamp ending in Z, such as 2026-12-01T14:00:00Z. If the configured time has already passed when the step runs, Atomic submits it for immediate publication. Choose a workflow start time early enough to prepare and review the post before its intended delivery. For Image, provide one image in Images. For Instagram / TikTok carousel, provide 2–10 ordered images. Select the matching Instagram or TikTok social account; TikTok also supports a single Image through Buffer. Each image has its own URL and alt text; use the slide controls to reorder the pair together. URLs must be permanent, publicly accessible HTTPS files. For video, use Public media URL and review any Instagram format, YouTube title/category, or other destination-specific fields shown. For TikTok photos, use Publish via Buffer and a connected TikTok account. Narrareach’s TikTok video destination does not replace this photo path. If the account is absent from the picker, review the Buffer connection before running the workflow. A network listed in a guide is not proof that your account is connected. The publishing screenshots use synthetic local connections and example media; they do not show a live provider submission. You can map an entire image array from a workflow input or previous step instead of entering each image manually. Each array item contains url and altText; keep descriptions aligned with the image order. When creating this workflow input, choose an array of objects. Review the resolved input before running it. Scheduling the workflow itself and setting a Buffer publication time are separate actions. Inspect the publishing result and Buffer queue afterward. If the result asks you to reconcile an uncertain request, check the existing provider item before retrying; an API timeout does not prove that nothing was submitted.

Publish Substack articles through Narrareach

Connect and verify the intended publication in Narrareach first, using an account with article-publishing access. In Atomic, add Publish via Narrareach, select its connection, and choose Substack (publication handle or URL required) as the publishing destination.

1. Enter the connected publication’s exact handle or URL in Substack publication. The destination option identifies the platform; it does not discover or choose a default publication for you.
2. Choose Article, then supply the approved title and HTML body. Review any subtitle and cover image.
3. Set Article publishing time to a future ISO timestamp with an explicit offset and set Article timezone to a named timezone, such as Europe/Belgrade. The default timezone is UTC. Leaving the time blank queues the article one minute after the step executes.
4. Enable Send Substack newsletter email only when subscribers should receive an email. It defaults off.
5. Enable Paid Substack article only for paid access. A native <hr data-type="paywall"> divider in the HTML sets the free preview boundary and also implies paid access at Narrareach.

The current article flow uses the connected personal Narrareach account. Team articles and short-form Substack Notes are outside this flow, even though a Note can be drafted in a Grid. An accepted result means the provider accepted the queue request. Publishing history remains PROCESSING while delivery is pending; confirm the final state and public URL in Narrareach. If a request times out, lacks a schedule identifier, or needs verification, inspect the existing queue and complete any verification in Narrareach before taking retry action. Do not submit a second copy merely because the final URL is not yet available.

Use an approved Grid revision

For a step offering Approved grid post cell ID (optional), supply that ID with Approved post revision. Atomic reads the saved approved draft, overriding manually mapped content fields. A newer or unapproved revision cannot publish, and the selected destination must match its platform. Keep the connection and delivery time explicit. See Social posts for review states, permissions, carousel editing, and revisions.

Human review steps

Use Human review when a workflow should pause for a person to approve a value or choose from generated options. For list reviews, give the output a clear variable name. When the list is an array of JSON objects, choose the preview field reviewers should see in each row. Full JSON remains available in review details. Keep the full selected object as the downstream output unless the next step only needs one nested value.

Publish a Webflow site

Use Get Webflow publishing domains to read the connected site’s production domains and publication state. Then use Publish Webflow site only after reviewing the whole site’s staged changes. This publishes all staged changes, including unrelated edits, to the selected production domains. It is broader than publishing one CMS item.

1. Select the same Webflow connection in both steps. Its API token needs sites:read and sites:write for site publication.
2. Review the returned customDomains. Put the intended domain IDs into Production domain IDs as a JSON array.
3. Map the returned publicationToken into Original publication token. If the state has changed, read it again and review before proceeding.
4. Keep a human review before the publish step. Confirm the selected domains and every staged change that will go live.

An accepted result means Webflow acknowledged the request; it is not proof of publication. Check published and outcome, then verify the rendered changes on each destination. An unknown result needs inspection before starting another run. The workflow assigns a stable publication identity within a run; a new run can create a new publication attempt. Webflow allows one successful site publication per minute. For design edits, first connect design access. Get Webflow design reads the current design; Update Webflow design saves changes without publishing. Changes to shared classes affect every use on the site. Review the saved design before the separate site-publication step.

Manage existing Webflow CMS content

Use the Content & Publishing steps to find and update an existing CMS item. Select the project-owned Webflow connection first, then the collection and item. This flow handles primary-locale CMS posts and collection pages. For an update, enter JSON using the collection’s native field slugs:

• CMS changes: only the fields you intend to change, such as {"name":"Updated guide title"}.
• Original values: the original values of exactly those fields from Get Webflow CMS content, such as {"name":"Previous guide title"}. This protects changes made by someone else since the read.
• Publication: choose Save draft or Publish live. Saving a draft leaves an existing live version unchanged. Publishing affects this CMS item, not the whole site.

If the result reports a conflict, read the item again and review the newer values before trying another update. Check saved, published, and outcome in the output: a saved draft is not a verified live publication. For a partial or unknown outcome, inspect the saved and live item with the verification step before retrying. Keep the attempt ID from the result when investigating an incomplete update. The screenshot uses illustrative local CMS data; it does not show a publication to an external site.

Programmatic page workflows

Programmatic workflow blocks are reusable. You can combine only the steps needed for the job:

• Detect programmatic page clusters: find repeated competitor URL patterns.
• Programmatic market research: combine the description with the current project and brand kit, identify direct competitors and similar projects, inspect how they implement the requested page pattern, and return one typed pageStructure. The structure includes the recommended page type, WDF/IDF terms, source URLs, layout requirements, and observed content-length ranges.
• Extract programmatic page structure: inspect a urls array and return a reusable page generation schema, shared section structure, and full-page screenshot URLs for each analyzed example. Competitor alternative pages include named sections such as hero, trust bar, comparison table, differentiator cards, testimonials, pricing, FAQ, and final CTA.
• Get Webflow collection structure: read fields from an existing Webflow CMS collection.
• Create programmatic collection structure: create or reuse a Webflow CMS collection from an extracted page structure or existing collection structure.
• Generate structured page content: generate one CMS-ready page object from a writing prompt and pageStructure. The step runs visible Research and Writing phases, validates required CMS fields and content-length ranges, then returns generated_page.
• Publish mapped content: map workflow values into WordPress or Webflow and publish or save drafts. For Webflow, use Map fields to fill fields one by one, or Paste JSON to pass the whole generated_page object.

Map WordPress custom fields

When you select a WordPress connection in Publish mapped content, Atomic reads the ACF fields exposed for the selected post or page type. Use Map fields to map workflow variables or fixed values one by one. Use Paste JSON to provide the complete ACF values object from one workflow variable or a pasted JSON object. Standard WordPress fields such as title, content, excerpt, slug, author, categories, and tags remain separate mappings. Author accepts a WordPress user ID or an exact username, slug, or display name. Use Get content from editor before publishing an existing Atomic content item. It returns one canonical field for each saved value: title, fullDescription, metaDescription, focusKeyword, featuredImageUrl, canonicalUrl, slug, content, wordCount, readTimeMinutes, status and score fields, linked record IDs, outline and NLP data, content structure, and timestamps. fullDescription preserves the complete summary; metaDescription is the short SEO version and is limited to 80 characters, ending in ... when it must be shortened. Read time is rounded up at 200 words per minute. The slug comes from the canonical URL path when available and otherwise from the title. If the custom fields do not appear, open the field group in WordPress and enable Show in REST API. Atomic sends the mapped values in the same WordPress publish request as the post or page. Use Programmatic market research when you need to discover what the market expects before creating a collection. Pass the same pageStructure into both Create programmatic collection structure and Generate structured page content. Use Extract programmatic page structure when you already know which URLs should define the pattern. Use Get Webflow collection structure when you already have a Webflow collection and only need more pages for it. For an existing Webflow collection, you can optionally select a reference CMS item in Get Webflow collection structure. Atomic uses the selected item’s field values as bounded examples so generation understands each field’s intended role, format, and level of detail. Reference values guide structure only; competitor-specific facts and claims are not copied into the new item. During Generate structured page content, the Research phase searches the selected project knowledge bases first. It then searches the live web and reads the strongest source pages needed to verify competitor and comparison claims. The Writing phase receives that evidence brief and the Webflow schema, but has no research tools; it produces the final CMS field values without inventing unsupported facts. Open Agent view on a completed workflow run to inspect the separate Research and Writing conversations. Agents can start the same project-aware capability with the Programmatic Market Research tool (programmatic_market_research). It returns a researchId immediately so long-running live research does not depend on one MCP request staying open. Results typically take around 10 minutes, which is expected. Call wait_for_programmatic_market_research to wait for completion, or get_programmatic_market_research to check the current status and retrieve the canonical pageStructure. The workflow step remains self-contained and returns the completed market-research structure directly. For Webflow collections, pageStructure preserves the exact Webflow type for every field. Short scalar copy should stay PlainText. Repeated blocks, formatted lists, tables, embeds, and multi-paragraph content should be RichText so collection creation, generation, and publishing all use the same type. Generate structured page content accepts only prompt and pageStructure. Put the audience, topic, CTA, constraints, and known facts in the prompt. The step fails instead of silently accepting unknown fields, missing required fields, or values outside researched content-length ranges. The Webflow programmatic collection structure template starts with a collection idea and sample-page prompt. It researches the market, creates the matching Webflow collection from the returned typed structure, generates a sample CMS page, and exposes both the research and generated object in the workflow output. Create programmatic collection structure also returns a Webflow wireframe image. Use wireframeImageUrl or wireframeMarkdown to review the intended Collection Page layout, including visible hero/body bindings and SEO metadata fields that should be configured in Webflow page settings instead of placed as text blocks on the canvas. Before running this step, open the Webflow connection in Settings → Integrations → Manage and choose Connect design access. Authorize the same Webflow site already selected in Atomic. The step now installs the Collection Page layout directly through Webflow’s official MCP. It checks the saved template root, heading hierarchy, and every required CMS binding before reporting layout_applied. A failed or unsupported binding reports layout_failed; a ready CMS schema alone does not mean the layout is complete. The generated wireframe represents the intended layout, so review the actual saved design before publishing. Use Get Webflow design to inspect a page, its styles, and current native tool schemas. Update Webflow design saves one native action using the fresh designToken and a stable UUID attemptId. Unknown outcomes require inspection; retrying the same attempt never repeats the write. Shared class changes affect every use on the site. Design edits do not publish the site. Visual snapshots require Webflow’s MCP Bridge App to be open in the Designer. When a generated page already matches the Webflow collection fields, select Paste JSON and use {{generateStructuredPageContent.generated_page}}.

Map WordPress SEO fields

In Publish mapped content, select the WordPress connection and content type, then choose Map fields. Below the ACF fields, Atomic shows writable SEO fields for the plugins exposed by that site’s REST schema: Yoast, Rank Math, or Squirrly. Map the fields shown for your destination, such as SEO title, meta description, or focus keywords; available fields depend on the installed plugin and connector capabilities. Install and activate the Atomic AGI connector and the intended SEO plugin on the WordPress site. If a plugin section is missing, check the connector version and the site’s exposed REST fields before running the workflow. ACF’s Show in REST API setting controls ACF fields; it does not enable an SEO plugin. Read-only plugin fields are not offered for mapping. Review the mapped values before publishing, then check the saved metadata in WordPress. The local example illustrates field mapping; no external publication was performed.

Schedule a mapped publication

In Publish content or Publish mapped content, choose Schedule (WordPress) and enter Scheduled date and time (UTC). Convert your intended local publication time to UTC first. A mapped variable must contain a UTC ISO date and time ending in Z, such as 2026-10-20T09:00:00Z. This schedules publication of the content, separately from when the workflow itself runs.

• WordPress: Atomic creates the remote post immediately with a future publication time. Check the scheduled item in WordPress after the run.
• Webflow: this API flow rejects future publication before creating the item. Use draft or immediate publishing here, then schedule in Webflow when needed.

Workflow run tab (inputs and output)

This is the control panel for a specific workflow:

• Set run inputs (for example, Current Date).
• Confirm brand context before run.
• Click Run workflow to execute.
• Read final output in the right panel.

Use this view when you need a clean rerun with controlled inputs.

Execution tab (live step progress)

Execution helps you debug in real time:

• Left side shows each step and status.
• Right side shows details for the selected step.
• Running steps reveal where time is spent.
• Click Cancel run to stop the selected active run without canceling the workflow itself.

Use this tab to find bottlenecks, stalled steps, or weak step prompts.

Run progress state

During a live run, watch for:

• Step stuck in running too long.
• Empty or partial step output.
• Unexpected transitions between steps.

If one step repeatedly slows or fails, simplify that step before scaling schedule.

Workflow history tab (single workflow)

Use this tab to review reliability over time for one workflow:

• Compare run outcomes.
• Confirm stopped runs show canceled.
• Track AI credit trend.
• Validate cadence against business needs.
• Re-open older runs for QA checks.

This is the best view for per-workflow quality review.

Schedule a workflow (command action)

From workflow detail actions, open Schedule workflow and define cadence:

• Daily
• Weekly
• Bi-weekly
• Monthly
• Inactive

Set schedule only after the workflow is stable in manual runs.

What to automate first

Start with tasks that are frequent and rules-based:

• Daily or weekly priorities for SEO and content.
• Content refresh candidate detection.
• Alerting for major movement or failures.
• Structured summaries for stakeholders.

Keep one-off strategy prompts in New chat.

What to fix first

If runs are unstable, fix in this order:

1. Inputs (wrong or missing context)
2. Step prompt clarity
3. Step order and dependencies
4. Schedule frequency

Most failures come from bad inputs or vague step instructions.

Weekly workflow checklist

1. Review failed runs.
2. Check output quality for completed runs.
3. Improve one weak step in the top-used workflow.
4. Remove duplicate workflows and keep one source of truth.
5. Reconfirm schedule only for proven workflows.

Keep in mind

• Small reliable workflows beat large fragile workflows.
• Naming matters for discoverability and adoption.
• A successful status does not always mean useful output.
• Schedule is an operations tool, not a quality fix.

Where to go next

• Agents: assign the right specialist to each step
• Automations: schedule recurring agent conversations around your workflows
• Tasks: track follow-through after workflow output
• New chat: prototype tasks before turning them into workflows