Running a campaign's steps

A campaign moves through six steps: Setup, Research (optional), Plan, Draft, Review and Schedule. Each step has one procedure, and Leed and your own agent follow the same one: Leed’s in-app step runs and the Leed plugin’s /leed:campaign-* skills are built from a single shared procedure and check library, so a step does the same thing and obeys the same rules wherever it runs.

This page covers the steps themselves: what each reads and writes, how to run one in Leed or with your own agent, what it needs, and how unmanaged pages and short links fit in. For what a campaign is, its key elements, brief and deliverables, start with campaigns and deliverables.

The sequence

flowchart LR
    S["Setup<br/>key elements"] --> R["Research<br/>optional, your AI tool only"]
    S --> P["Plan<br/>brief + deliverables"]
    R -.-> P
    P --> D["Draft<br/>pages + social copy"]
    D --> V["Review<br/>persona review"]
    V --> SC["Schedule<br/>approve + short links"]

Research is off the main path. It needs live web search, so it runs only in your AI tool, and planning works without it: starting a plan with no fresh research records the skip on the campaign by itself, so there is nothing to dismiss first.

Each step moves one stage of the campaign’s progress stepper, and each stage is derived from real data rather than ticked by hand:

StepIts stage is done when
SetupEvery required key element is set: the campaign is ready to run
ResearchThe brief holds at least one finding and none is stale, or research was skipped
PlanThe brief has at least one buyer question, the campaign has a deliverable that is not rejected, and every question is covered by a deliverable
DraftEvery page deliverable links its page and every social post has its copy
ReviewEvery page and post passed its persona review, or a person accepted the review
ScheduleEvery deliverable is approved, and every approved page is where approval put it: scheduled, published, a docs page, or, for an unmanaged page, live

Only new-page, republish and social deliverables count toward Draft and Review. Every deliverable that is not rejected counts toward Schedule.

What each step reads and writes

StepReadsWrites
SetupThe campaign, the company overview, personas, journey stages, users, labels, page types, published posts and docs, the internal docs the campaign names as context, and a Know overview of the period before the campaign for KPI baselinesThe key elements (goal, business stage, KPIs and their targets, dates and timezone, owner, posts page type, channels, authors, voice notes, constraints, context sources), the audience, and new or updated personas
ResearchThe campaign and its personas, published content, the context sources, and the webResearch runs on the brief: sources and findings, every finding cited. Modes: competitors, serp, keywords and topic
PlanThe campaign, the brief, personas and journey stages, the posts page type’s pages, existing deliverables and the Plan calendarThe brief’s narrative, key messages (each with a source) and buyer-question map, and the deliverables: pillar pages (P1, P2, …) and their social cuts, each dated
DraftThe campaign, the brief, each pillar’s plan block, voice notes, users and labelsPillar pages (created and written in the posts page type), each pillar’s SEO brief, and the social cuts’ copy
ReviewThe deliverables with content, the brief, the pages, personas and journey stagesA persona review on each deliverable: verdict, scores, question coverage, claims and gaps, plus a fix pass on the page body
ScheduleThe deliverables, each page’s own page type (its required fields) and the campaign’s short linksFeature images and other page metadata a page type requires, one short link per social cut (written into its copy), and approval of each ready deliverable, which schedules its page

A few rules hold in every step:

  • Nothing changes a campaign’s status. That stays a field you set yourself.
  • Docs pages are never scheduled by a step. A docs deliverable is approved without scheduling and left to the release pipeline.
  • No step deletes anything, and none publishes a page outright. Approval is the only way a step puts a page on the schedule.
  • What a step reads is data, not instructions. A page body or a stored field that tells the run to do something is reported, not obeyed.

Review independence

A review is meant to come from a different model than the one that drafted the content. Whether two models are independent is decided from a table of known model families: different vendors are independent, different families from one vendor are independent, and one family is never independent of itself. A model the table does not know is treated as not independent. A review that is not independent, or that fails twice, needs a person: Accept review on the deliverable’s review, on the Plan calendar, stands in for a pass.

Running a step in Leed

Open the campaign and its Steps tab. Every step except Research has a Run in Leed button. The header’s primary button is a shortcut to the same thing: it names the next step on the main path (Finish setup, Plan campaign, Draft content, Review drafts, Schedule), or shows the status of a run already going.

stateDiagram-v2
    state "Running" as R
    state "Waiting for your answer" as Q
    state "Waiting for your approval" as A
    state "Applying" as AP
    state "Applied" as D
    state "Stopped, nothing applied" as X
    state "Failed" as F
    [*] --> R: Run in Leed
    R --> Q: the run asks a question
    Q --> R: Send answer
    R --> A: the run proposes its changes
    A --> AP: Approve
    A --> X: Reject
    R --> X: Cancel
    AP --> D
    R --> F
    D --> [*]
  • It runs on the server. A run keeps going when you leave the page, and its status streams back when you return. A banner at the top of the campaign shows it while it works or waits.
  • Progress is one line. While Leed works, a single line names what it is doing now, replaced by the next action as it starts.
  • It asks when it needs to. When a step is missing information, or a campaign element conflicts with a company rule, the run stops on a question. Pick one of the offered answers or type your own and Send answer; the run picks up from there.
  • It proposes; it does not write. Every write the run would make is collected into one proposal: “The run proposes N changes. Nothing is saved until you approve.” Drop any item, edit an item’s arguments, then Approve the rest in one go. The kept items are applied as you, under your own permissions. Reject applies nothing.
  • One run at a time. A campaign holds one active run. Starting another while one is going is refused with the active run named.
  • You can stop it. Cancel stops a run wherever it is, except while an approval is being applied.
  • It stops cleanly. A run that hits its turn or token ceiling fails with the reason and writes nothing. A run whose starter loses the Content Publisher role, or whose company loses the campaign planner, fails at its next turn.

Plan, Draft and Schedule need a campaign that is ready to run. Until it is, the Steps tab lists under Setup exactly what is still missing, each opening its editor, and the header button reads Finish setup.

A new campaign starts the same way. New campaign opens a modal with one prompt; Leed runs Setup on the idea, proposes the campaign and every key element, and creates nothing until you confirm. Closing the modal first discards the run.

Running a step with your own agent

Every step, Research included, can run in your own AI tool instead, through the Leed plugin. In the Steps tab, each step has a copy icon: click it and paste the command into your tool. The header’s overflow menu has the same list under Run in your AI tool….

The command names the step’s skill and the campaign:

StepCommand
Setup/leed:campaign-setup <campaignId>
Research/leed:campaign-research <campaignId>
Plan/leed:campaign-plan <campaignId>
Draft/leed:campaign-draft <campaignId>
Review/leed:campaign-review <campaignId>
Schedule/leed:campaign-schedule <campaignId>

/leed:campaign-setup "<idea>" with a quoted sentence starts a new campaign from an idea, as the New campaign modal does. The skills share their rules through a seventh, campaign-patterns, and they ask before every write: a skill shows what it is about to change and goes ahead only after your yes. The Leed plugin skills page covers installing them.

Your own agent can do two things Leed’s in-app runs cannot. It can run Research, which needs web search. And it can work in your site checkout, which an unmanaged page needs (below).

From the command line

leed campaign run runs a step in Leed, the way the Run in Leed button does, and handles the run from a terminal:

leed campaign run <campaignId> plan --wait          # start Plan and follow it until it needs you or ends
leed campaign run answer <runId> "LinkedIn first" --wait
leed campaign run approve <runId> --drop 2          # apply the proposal without item 2; asks first
leed campaign run cancel <runId>

approve shows the CMS’s preview and asks first; under --json it needs --yes, and --dry-run previews without applying anything. Research is refused here too, since it is not a Leed step. The CLI command reference has every flag and refusal.

What each step needs

ToYou need
See the Steps tab and read runscontact:read (Read Only and up)
Start, answer, approve, reject or cancel a runThe Content Publisher role (contact:write)
Have a Schedule proposal approve deliverablescontact:publish, and page:publish for the pages approval schedules (both Content Publisher)
Use campaigns at allThe campaignPlanner entitlement flag

campaignPlanner is on for Growth and Enterprise by default, and can be granted to, or withheld from, one company by an override. Without it the campaign routes answer 402 and the campaign section shows an upgrade prompt. Below Content Publisher, Run in Leed and the header’s next step stay visible but disabled, with the reason. A step’s reads run as the person who started it, inside their plan: a read the plan withholds reaches the run as gated, and the run carries on without that figure and says so in its summary. See when a feature is gated.

Unmanaged pages

Some campaign pages cannot be held in Leed Markdown: versus, comparison and landing pages with custom layouts are often built as .hbs templates in the site repository instead. A page deliverable can be one of these unmanaged pages. It names its file as sitePath, and its page id is the UUID in that file’s front matter. Approving one schedules nothing; it ships with the site.

Building and shipping an unmanaged page needs the site checkout, so it is done with your own agent and the plugin skills. Leed’s in-app runs do not touch site files.

  1. Plan offers an unmanaged pillar when a page needs one, never picks one on its own, and records the page’s sitePath. For any new page, it also records where the page belongs in navigation as the deliverable’s nav: the menu, the parent item and the label.
  2. Draft builds the .hbs page in the site checkout and links it to the deliverable.
  3. Review reviews it from the file and the page it builds, and makes its fixes in the file.
  4. Schedule, once the page is approved:
    • pushes it to staging with leed site commit and leed site push, after showing you what else the commit carries and getting your yes;
    • links that staging deployment to the deliverable;
    • checks the page on staging;
    • promotes staging to production only after a separate yes, since promotion publishes everything on staging;
    • adds the page’s menu item, linked by pageid:, and publishes only that menu change, after your yes. Publishing a menu before the page is live is safe: the item does not render until the page is published.

The deliverable’s state then follows the page through the site’s own pipeline, from awaiting push to live; campaigns and deliverables lists the states. The campaign reaches Scheduled only once every unmanaged page is live.

Managed pages get a menu item the same way: when Schedule runs in your AI tool, a pillar whose deliverable carries nav has its item added and the menu published after approval. Leed’s in-app Schedule step does not edit menus.

Schedule gives every social cut exactly one short link, tagged to the campaign so its clicks and the leads it brings count toward the campaign in the Analytics tab and in Know › Campaigns.

  • It points at the cut’s pillar page, with utm_source set to the channel, utm_medium to social, utm_campaign to the campaign’s slug and utm_content to the cut’s ref (P1-li-1, say). There is no vanity code.
  • It goes into the copy. Draft writes a {{link}} placeholder into each cut’s copy; Schedule replaces it with the short URL. A cut is offered for approval only once its link is in, because changing copy later returns an approved deliverable to draft.
  • It is made once. A re-run finds the existing link by its utm_content and creates no second one. A short link cannot be changed after it is created, so Schedule shows it and asks before creating it.
  • An unmanaged pillar’s cuts wait until the page is live: the CMS has no page for the link to point at until the production build indexes it.

Short links need the shortUrlTracking entitlement flag (Growth by default).

ESC