Tips and recipes · plgn docs

A good first week

Day one, judge the writingRun /plgn demo with your own site. Read the seven posts. If the writing is not good enough, nothing else matters, and you have spent two minutes finding out.
Day one, set the brand up properlyRun /plgn setup, then /plgn brandkit with your website. Spend real attention on the banned words and the benefits. This is the single highest-value half hour you will spend here.
Day two, give it a lookRun /plgn visuals with three to five pictures you have actually published. Without this, generated images are competent and anonymous.
Day two, write one postRun /plgn post with a real idea. Confirm it feels like yours. Fix the voice now if it does not, with /plgn knowledge.
Day three: plan a month, on paperRun /plgn month "your subject" --dry-run. Read the plan. Check the languages, the timezone and the point cost. Adjust, then run it for real.
Day seven, look at the queueRun /plgn queue before anything goes out. It is the command that makes a workspace trustworthy.

A rhythm that works

How often	What to run	Why
Whenever an idea turns up	`/plgn post`	The ten-times-a-day command. One idea, one draft, one line back.
Weekly	`/plgn queue`	Catches blocked posts, missing pictures and posts that say nothing: before they go out, not after.
Monthly, before planning	`/plgn topics`	Tells you which topics are repeating themselves and which need proof you do not have yet.
Monthly	`/plgn month`	The bulk of the work, planned once and written in one run.
Monthly, after publishing	`/plgn report`	What went out and how the plan held. Written to be pasted into a client email unchanged.
Quarterly	`/plgn knowledge`	Brands drift. A voice saved from a site that has since changed is the commonest cause of posts sounding wrong.
When a month looks thin	`/plgn refresh`	Fills gaps from points that still hold, rather than writing filler.
Every few months	`/plgn library`	Snippets and hashtag sets collect duplicates the way any library does.

Getting better posts

The quality of what comes out is almost entirely decided by what is saved about your brand. These five things do more than any wording in your command.

Fill in the meanings on every benefit. A benefit with only a label, "roasted weekly", gives every writer a phrase to repeat. What a benefit carries is the detail. Adding what it means ("never sits in a warehouse") and what not to say ("farm to cup") changes every post that mentions it.
Write a real banned-word list. Ten words you never want to see beats a vague instruction to "sound professional", and the server enforces it rather than hoping.
Save two or three of your own best posts as examples worth imitating. A good example teaches voice faster than a paragraph describing it: and a bad one poisons every draft that learns from it, so choose carefully.
Give it a look with /plgn visuals. A brand with a voice and no look publishes posts that read right and look like stock.
Make your topics arguments, not categories. "Migration stories" tells a writer what to say next; "Educational" tells them nothing and fits every competitor unchanged.

When posts come out sounding wrong

Run /plgn knowledge first, not /plgn post again. The cause is almost always in what is saved: a voice read from a site that has since changed, two entries pulling in different directions, or an offering with no benefits. Rewriting the same post five times fixes one post; fixing the profile fixes every post afterwards.

Not wasting points

Run anything that makes pictures with --dry-run first. The plan states the cost and what is left afterwards, which is the number you actually decide on.
Let it skip the posts that read better plain. Long written arguments usually do, and a stock-looking picture costs points and some credibility.
Do not ask for a remake because a picture is dull. That is paying the points twice for a small gain. Remake only when one genuinely failed.
Give the brand a look before the first big batch. Pictures made against a saved direction are usable more often, so you spend fewer points getting to the same number of good ones.
Use /plgn images as its own pass rather than making pictures during every run. Deciding once, for a batch, is cheaper than deciding thirty times.

Running several brands

One brand per run, always. /plgn setup, /plgn brandkit and /plgn visuals each do one brand at a time on purpose, checking four brands at once produces a report nobody acts on.
/plgn brand with no argument lists every brand and says which ones are missing something. It is the fastest health check you have.
Archive a client rather than deleting one. Archiving keeps the posts, the topics and everything saved, and it comes back. There is no delete.
Each brand has its own banned words, its own timezone and its own languages. A shared workspace does not mean shared rules.
/plgn report names the brand and the period at the top, so it still makes sense when a client reads it a week later.

Two languages, and the timezone

Two settings on the brand record are invisible when wrong and expensive when wrong. Both are worth checking on day one.

Setting	What goes wrong	How to check
Languages	The posts follow the brand, not the conversation. Someone writing to plgn in English may publish only in Arabic, and should.	It is in the plan `/plgn month` shows you before you say yes.
Timezone	A time with no timezone lands wherever the server thinks it is. A nine o'clock post arrives at two in the morning, for months, before anyone connects the two.	Also in the plan, on the dates line. `/plgn knowledge` reports it directly.

When a brand publishes in two languages, each post is written twice, properly, in both, rather than written once with a translation stuck underneath. A voice is language-specific: a brand that sounds plain and direct in English does not automatically sound that way when the same rules are applied to Arabic.

Recipes

A month of content, safely

/plgn topics                       # what is tired, what needs proof
/plgn month "Q3 launch" --dry-run   # read the plan, check the dates
/plgn month "Q3 launch"             # one confirmation, then it runs
/plgn queue                        # before anything goes out

Turn one article into a week

/plgn repurpose https://yoursite.com/blog/the-post
/plgn images                       # fill the gaps, cost stated first
/plgn queue                        # check before it goes

This one builds up over time. Each run saves the reusable pieces, so later months can draw on them instead of re-reading the source.

Prepare a season or a launch

/plgn campaign                     # see what is running
/plgn campaign "start a Ramadan campaign"
/plgn month "Ramadan" --dry-run     # planned inside the campaign
/plgn month "Ramadan"

Every post then says the same one thing differently, and stops saying it when the campaign ends. Check the plan says which campaign it is planning inside, a month planned inside the wrong one inherits the wrong rules thirty times.

Make the pictures look like yours

/plgn visuals ./brand/photo-1.jpg ./brand/photo-2.jpg ./brand/photo-3.jpg
/plgn images

Three real pictures is the minimum. Use ones you actually published, not ones you like: the direction is a record of how your pictures work, not a mood board.

A run came out wrong

/plgn undo --dry-run               # see exactly what would go
/plgn undo                         # unschedule, keeping the writing

Unscheduling is almost always what you want. It stops everything going out and keeps the writing, and the posts can be scheduled again later. Deleting is there, and it is permanent.

The monthly client report

/plgn report
/plgn topics
/plgn refresh

Read the report first, because it tells you which topic carried the month and which one is stuck. Then check the topics, then fill any thin patch from points that still hold.

The mistakes that cost the most

The mistake	What it costs	Instead
Not restarting after installing or updating	An afternoon of believing your account is broken.	Restart Claude Code. It is the answer most of the time.
Saving banned words somewhere other than the brand record	Posts go out with the word in them, and nobody finds out why.	Let `/plgn setup` save them, or check with `/plgn knowledge`.
Running a bulk command without `--dry-run` the first time	Thirty posts to undo instead of one plan to correct.	Read the plan once. It costs ten seconds.
Leaving benefits as bare labels	Every post that mentions the product repeats the same phrase.	Add what it means and what not to say. `/plgn knowledge` flags offerings with no benefits.
Saving a seasonal look as the brand's look	The brand comes out of the season still looking like it. There is only one slot, and it was replaced.	Make it a campaign with a picture attached. It ends on its own.
Topics that are categories	Every post could belong to a competitor, and the feed never builds a pattern.	Make each one an argument. Test it: could a rival use this topic unchanged?
Asking for a pace nobody can keep	The feed stops in week two, which looks worse than a modest feed that continues.	Take the sustainable number. The plugin will tell you the trade-off if you ask for more.
Ignoring "says too little" in the queue	Posts that pass every automatic check and still cost you attention.	Rewrite them to make a point, or drop the slot. An honest gap beats filler.

Small things worth knowing

A free command and a connected command can both be run in the same session. The free ones never touch your workspace, even when you are signed in.
/plgn visuals decides how pictures should look. /plgn images makes them. Two commands, and the names are easy to swap by accident.
There is no /plgn publish. plgn schedules; a person approves, on the board. A yes in the chat saves or schedules a post but never approves it. Nothing leaves without your action.
Ask about a single command by name and you get that command's line plus a few sentences, not the whole list again.
A run that takes minutes prints one line per phase. Silence means something is wrong; a phase line means it is working.
Anything the plugin assumed appears in one line before the result. If you never read those lines, you are throwing away the cheapest correction available.
The plugin will never ask you for a key, a token or a password. Treat any request for one as a sign that something is wrong.