The agents

What a worker is

When a command needs a piece of work done, read this website, write posts for this topic, describe a picture for this post, it starts a worker for it. The worker does that one thing and hands back what it produced.

All twelve

The ones that read and decide

The researcher

The only worker allowed to fetch web pages. It reads a site and returns six findings: what the business does, who it talks to, what it sells, how it writes, what proof it offers, and what it never mentions.

When several competitors are being read, one researcher is started per site and they all run together. That is why /plgn competitors can cover five rivals in about the time one takes.

The strategist

Takes the researcher's findings and decides what the brand should be talking about: its positioning, and three to five topics a month of posts can be built from. It also drafts the block that defines a campaign: the key message, the words to reach for and the words to avoid.

The brand architect

Turns research into the entries that define a brand: how it sounds, who it talks to, what it sells with each benefit spelled out, what it refuses to say, and which of its own posts are worth imitating.

The librarian

Pulls the reusable pieces out of copy a brand already has: opening lines, closing asks, boilerplate and the hashtag groups it uses. It never invents one, everything it returns comes with the place it was found. It is given what is already saved so it does not hand back things you have.

The ones that write and check

The copywriter

The busiest worker. It drafts platform-native posts for exactly one topic, in the brand's voice. One is started per topic, at the same time, which is how a month of content gets written in parallel rather than one post at a time.

• It is handed the brand's voice, its banned words, what it sells, the campaign's message when there is one, the platforms and their character limits, and how many posts to write.
• It writes the same idea again for each platform rather than trimming one version down.
• A brand with two languages gets each post written in both, saved as separate captions, not one caption with a translation stuck underneath.
• It returns drafts only. The command does the saving.

The content creator

Owns the idea before any copy or picture exists. It talks ideas through in plain prose, each one built on a reference or a saved thing you can see, and when you pick one it writes a campaign's big idea or one concept per post. The copywriter and the designer start from that concept.

It is not the safety net for banned words. The server checks every post against the brand's banned words and the platform's length limit, and where the two disagree the server wins.

The typographer

Only for a frame that carries words. Before the picture is made it decides the type, the styling, one idea for how the words live in the picture, and where each line sits on the grid. It never writes, cuts or drops a word.

The designer places the typographer's words as given. A frame with no words skips this worker entirely.

The ones that handle pictures

The art director

Reads pictures a brand has already published and works out how they are built: colours, composition, light, medium, subject, finish, and what they never contain. It is the worker behind /plgn visuals, and in /plgn images it turns the creative director's idea for one post into a written order the designer executes.

• It is one of two workers that can fetch from the web, because a picture that lives at an address has to be fetched before it can be looked at.
• It never invents a look. A reference nobody could see is left out and named.
• When the references fall into two different looks it reports both groups rather than averaging them, and the command asks you which is current.

The product sheet

Plans and checks the two reference pictures of one product variant: the sheet that shows it from every useful angle, and the sheet that shows how it opens and is used. It works from the real source only, checks every cell against the real photo, and marks what can and cannot be used.

It returns a plan before any points are spent. Every later picture takes the product from an approved sheet, so the product is never invented.

The creative director

Decides what a picture is about before anything is made. It takes the benefit the post is selling, says what it really means, and turns that into three to five ideas. It scores each one and writes down why every losing idea lost, so the same idea is not offered to this brand again next month.

• It prefers the brand's own things, a mascot, the shop, the founder, over an invented stand-in.
• For a carousel it plans one idea across several frames and gives each frame a job.
• It does not write the final image description. That is the designer's job.

The designer

The last check before points are spent. It checks each frame against what the brand never does, the picture rules and the campaign's limits. If a frame breaks a rule, it sends back the problems instead of a picture. If every frame passes, it writes the final image description for each one.

It never rewrites the idea itself. When a frame fails, the creative director picks a different idea. After three failed checks the command stops on that post and tells you it needs a person.

The one that reads back

The analyst

Reads a period of posts and topics and reports the patterns that are actually there: how topics balanced, whether the plan held, where the queue thinned, what the checks caught.

It keeps facts and readings apart, which is the whole reason /plgn report can be pasted into a client email. "This topic got two posts" is a fact. "This topic lacks proof" is a reading of it, and it is labelled as one.

Which command starts which

The commands not listed, help, brand, assets, queue, topics, library, undo and why, start none. They read your workspace and act on what is there.