X replies agent
How the X replies agent searches for posts worth answering, what each proposal carries, and how you post the reply in the thread.
What it does
The X replies agent works the daily engagement routine of the social skill. It answers the replies and mentions under your own posts first, then the accounts you engage with, then the day's posts where your product genuinely belongs, and drafts one reply to each. A reply answers what was said first; the product is mentioned only when it is the answer, and never as a pitch.
Each draft comes with one sentence saying why that post is worth answering. The post itself is carried as X served it, with its author, its text, its date, its numbers and its link, rather than as the model retyped it.
When it runs
It runs as a daily Loop, on from the moment the connection exists. The cadence comes from the agent and is not editable, and it carries no weekly cap; what bounds a day is the search budget and the rule that a thread is answered once, ever.
What it needs
The X connector. Without it the agent is unavailable and does not run. Writing your own posts is a separate agent, X posts, on its own switch, and the two share the same connection.
What it reads
The foundation doc, the growth plan and the product memory first: the positioning is what decides whether a post is one you belong under, and the brand voice is what the reply sounds like.
Then the replies and mentions of your X account, read from X's own list for it: X's search leaves out much of what new accounts write, so this is how the agent sees who answered you. X bills each post and author that read returns, and Kwern takes it from your plan's Usage.
Then the circle: the accounts your product engages with, which the agent keeps as a data output and rereads every run, so it builds on yesterday's list rather than starting over.
Then X, searched up to five times a day for this agent. Kwern reads through its search service and falls back to the official API only when that refuses; a day where X cannot be read leaves the agent nothing to answer. Every post a search returns is kept for the length of the run, and a reply can only be drafted for a post the run actually read.
What it leaves you
Proposals in one category, reply. Each reads in three parts: what the post said and what Kwern proposes beneath it, who wrote it, and why this post. X's API no longer lets Kwern reply to most posts, so there is no approve on them: you edit the reply if you want to, copy it, open the post from its link, post the reply from your own account and mark the proposal as done. Kwern reads no metrics back for a reply it did not post.
Beside the replies, proposals in two more categories, paced by the skill's daily routine: engagement, a like or a repost of a post, or a follow of its author, and outreach, a direct message to a new connection, written ready to paste. Kwern's X access cannot make any of them, so they carry no approve either: you open the post, or the author's profile for a follow or a message, do it from your own account and mark the proposal as done.
The skills it runs
It enters through community-marketing, and may activate social.
Editing its instructions
The agent's page in the app carries the instructions it is given at the start of every run, and you rewrite them there — what a post has to say before it deserves an answer, how long a reply should be, the conversations to stay out of. The cadence is not editable. The switch reads on, off or paused, where paused means Kwern stopped it for a reason and will resume it when that reason clears.
X Communities agent
How the X Communities agent finds the X Communities worth joining, writes a post for each one's members, and posts it into the Community once you approve it.
X posts agent
How the X posts agent listens before it writes, what it learns from your own posts, and how an approved post reaches your account.