x-post

---
name: x-post
description: >-
  Turn one raw idea into an X post: three hooks off 19 formulas, scored offline,
  one humanized draft counted in weighted characters, ready to copy and post by
  hand. Use when the user says "write a post about X", "tweet this", "give me a
  hook", "turn this into a post", or has an idea and no first line. Free account
  by default (280 weighted characters, link in the first reply). Never posts.
---

# x-post

One idea in, one post out. Two tools live beside this file and both run. Use
them; do not eyeball the hook or the length. Run them from this skill's folder.

```bash
python3 xcount.py draft.txt                 # weighted length, the way X counts
python3 postscore.py draft.txt              # hook, specificity, reply pull, link, bait
python3 postscore.py --hooks hooks.txt      # rank three candidate first lines
python3 postscore.py draft.txt --split-link  # POST and FIRST REPLY blocks, link moved out
```

On Windows, where this says `python3`, use `python` (`python3` there is usually the Microsoft Store
shortcut and fails with "Python was not found"; try `py` if `python` is also missing).

## Before you write

1. Read `~/.claude/x/voice.md` (or `~/.claude/x/<handle>/voice.md`). If it does
   not exist, ask for three of the user's own posts, infer the voice, and write
   the file from `templates/voice.md`. If that template is not on disk (a by-hand
   install copies only `skills/`), write the file with these headings: Who I am on
   X (handle, what I do, who I write for, free or Premium, where the money comes
   from), How I sound (sentence length, contractions, slang, capitalisation),
   Things I never say, Proof I can use (only facts the user gave), Off-platform
   path (the one link), Never (invent a number, ask for likes).
2. Default persona: a small account on the free tier, earning off-platform
   through clients, a product or a newsletter. 280 weighted characters. Do not
   assume Premium. Premium features are documented in `references/x-rules.md`,
   not tested by the author.
3. If the idea is thin, ask one batched question: what happened, to whom, what
   did it cost or return. A post needs one specific true thing.
4. **Never invent a number.** If the draft needs a figure the user did not give,
   write `{{your number}}` and say so. `postscore.py` blocks any draft that still
   contains a `{{...}}`.

## The loop

1. **Three hooks, not one.** Choose three formulas from `hooks.json` that fit the
   idea. Different formulas, not three rewrites. Put the first lines in a file.
2. **Score them.** `python3 postscore.py --hooks hooks.txt`. Show the ranking.
   If the best is under 50, you do not have the hook yet. Get a better fact from
   the user instead of editing.
3. **Write the draft** on the winning hook. One idea. Plain words. Fits 280
   weighted characters. A link in the root post is a hard stop on a free account:
   `postscore.py` caps the score at 49 and blocks the draft.
4. **Count and score it:** `xcount.py` then `postscore.py`. Fix every flag.
5. **Humanize it:** run `/x-human` before showing anything.
6. **Print the post block.** When the draft had a link, ALWAYS print a second block
   titled FIRST REPLY containing the link, and nothing else. `--split-link` prints both.

Real output (sample draft, free account):

```
POST SCORE  93.8  STRONG   (203/280 weighted)
----------------------------------------------------------------
  HOOK         ###################.  95.0  61 weighted characters
  SPECIFICITY  #################### 100.0  9 concrete markers in 42 words
  REPLY PULL   ###############.....  75.0  question present
  LINK         #################### 100.0  no link in body
  BAIT         #################### 100.0  none
```

A draft with a link in the root post is capped and blocked (real output):

```

POST SCORE  49.0  WEAK   (141/280 weighted)
score capped at 49: link in root post
----------------------------------------------------------------
  HOOK         ###################.  95.0  52 weighted characters
  ...
BLOCKING: Link in root post. On free accounts, links in the root post were observed to cost reach (observed, not coded). Put it in the first reply.
```

The fix is to move the link out. `--split-link` prints the two blocks to hand to the user:

```
POST (link removed)
I sent 400 cold DMs in 30 days and 2 became clients.

What is the first line you send to a stranger?

FIRST REPLY
Full breakdown:
https://example.com/dm-breakdown
```

A weak draft looks like this, and `postscore.py` exits non-zero on it:

```
POST SCORE  37.1  WEAK   (161/280 weighted)
  BAIT         ....................   0.0  like/repost solicitation, "RT/like if" bait, "follow for more"
               - engagement bait ... This is also an Original Content Rewards ineligibility problem
BLOCKING: engagement bait; hashtag wall (more than 2)
```

## Rules that matter

- **Link in the first reply, never the root post.** On a free account a link in the
  root post is a hard stop (score capped at 49, draft blocked). Label it honestly:
  *observed, not coded*. The open-sourced ranking parameters contain no link penalty. Reach loss
  on link posts from non-Premium accounts was observed in March 2025 (Buffer,
  18.8M posts). It is a measurement from outside, not a coded rule.
- **ALLOW_LINK.** If the user has a reason to keep a link in the root post (say so out loud: it
  is the one thing the scorer will not let through by default), run `postscore.py draft.txt
  --allow-link`. The hard stop becomes the old warning, the score is not capped, and the other
  checks still apply. Never use it on your own initiative.
- **No bait.** "Like and repost if", "follow for more", "bookmark this" are ranking
  risks and Original Content Rewards ineligibility problems. Cut them.
- **At most one hashtag.** Zero is normal.
- **The first line works alone.** No "A thread", no "Here's why", no "Let me explain".
- **Replies are cheap to earn with a real question.** End on something answerable
  in one line. "Thoughts?" is not that.
- **Weights scale predicted probabilities, not raw counts.** Never repeat the
  "one report cancels 468 likes" line. See `references/x-algorithm.md`.

## Fine print

`postscore.py` is five local heuristics. It does not know how a post will do.
Measured on a public tweet corpus, its score did not predict which tweets beat
their account's median likes (AUC 0.45 to 0.50). Treat the number as a lint
result, never as a forecast, and say so if the user reads it as one. It never says a post is "undetectable"
or "will go viral".

## Without the tools

If you cannot run the Python tools (for example you pasted only this file), do the checks by hand and
say so. Count weighted characters yourself: 280 on a free account, a URL counts 23, CJK characters and
most emoji count 2. Then check: the first line works alone; there is a specific number or event the
user gave; no "like and repost if" or "follow for more"; no unfilled `{{placeholder}}`; at most one
hashtag; the link is in a separate first-reply block. Tell the user the score was not run.

## Approval gate

1. Show the finished text in one fenced block, ready to copy. If there is a link,
   show it in a second block titled FIRST REPLY.
2. End the message with: Reply yes to log it.
3. On "yes", append one line to `~/.claude/x/log.md` (or the handle's folder):
   date, skill, first line. That is all "yes" does.
4. Never post, like, follow, reply, DM or scroll. You write; the human posts, by
   hand, on x.com.

• SKILL.md 7 KB
• hooks.json 8 KB
• postscore.py 17 KB
• xcount.py 7 KB