---
name: "x_growth_playbook"
description: "Grows an X (Twitter) account organically: algorithm-aware strategy, the reply game, thread craft, analytics loop. Use when the user asks how to grow on X, get more reach/followers, audit an account's performance, improve replies/threads, or plan an X content cadence."
metadata: { "includeInPrompt": true }
---

# X Growth Playbook

## Purpose

Run organic X growth as a repeatable loop: **diagnose → fix pillars → publish → measure weekly**.
Everything the algorithm side claims is grounded in X's open-sourced recommendation code
(Jan 2026, `xai-org/x-algorithm`); craft advice is practitioner opinion and labeled as such.

## Workflow

### 1. Diagnose the account (before touching anything)
1. Get the baseline: 28-day analytics (analytics.twitter.com, or Creator Studio → Analytics).
   Record: impressions, profile visits, followers gained, and per-post impressions/likes/reposts/replies.
2. Score it with `bin/x_engagement.py --csv <export.csv>` — it computes engagement rate and a
   weighted signal score per post. Rank posts by score, then read the actual words of the
   top 5 and bottom 5.
3. Classify the diagnosis (one or more):
   - **Distribution problem** — good content, low impressions (hook/algorithm/format issue).
   - **Resonance problem** — impressions but no replies/quotes (content doesn't spark conversation).
   - **Conversion problem** — impressions and engagement, no follows (weak profile, no clear follow reason).
   - **Cadence problem** — too little output to learn from, or erratic volume.

### 2. Fix the pillars (in this order)

**Profile (one-time, high leverage).** Clear photo or brand mark, a bio that states who you
serve and what you post about, a pinned post that is your best work. The pinned post is
what profile-clickers land on — make it earn the follow.

**Signal strategy (the algorithm core).** Details in `references/algorithm-signals.md`.
The two rules that drive everything:
1. Optimize for **replies that start conversations** and **shares/DMs**, not likes. Likes are
   the weakest positive signal in the model.
2. Avoid negative signals at all costs: nothing that gets reported, muted, or marked
   "not interested." That means no engagement bait, no reply spam, no controversy-for-attention.

**The reply game.** Details in `references/reply-game.md`. Replies to larger accounts are the
cheapest distribution an account can buy — but only when each reply could stand alone as a
good post. The era of reply rings is over (X's Head of Product has stated replies don't count
toward revenue sharing); the reply must add real value.

**Thread craft.** Details in `references/thread-craft.md`. The hook (first post) is the whole
thread: if the first post doesn't earn a stop, the rest never gets read.

### 3. Run a weekly cadence (opinion, not algorithm fact)
- Post at a pace you can sustain indefinitely; consistency beats bursts. A new or small
  account should aim for daily output so there's something to measure each week.
- Mix formats: standalone posts, replies, threads, and media. Never publish external links as
  the whole post — lead with the value in the post itself (analyst consensus: external links
  are suppressed).
- Engage daily: reply to others, answer every real reply on your posts with substance.

### 4. Measure weekly (the loop that compounds)
Read `references/analytics-loop.md`. Each week: export the CSV, run `bin/x_engagement.py`,
log the per-week totals, and write 3 sentences: what worked, what flopped, what to try next.
Kill the bottom 20% of formats/topics; double down on the top 20%.

## Output Contract

When asked to **audit** an account or **plan** its growth, deliver:
1. **Diagnosis** — which problem(s) from step 1, with numbers.
2. **Top 5 / bottom 5 posts** — post URL, one-line reason it worked or didn't.
3. **Action list** — ≤7 concrete actions, ordered by expected leverage, each with a
   "done looks like" line.
4. **7-day content plan** — topics × formats, one line each, matched to the account's voice.
5. **Weekly review template** — what to measure, when.

## Operating Rules
1. **Honesty about provenance.** Verified algorithm facts ≠ analyst estimates ≠ practitioner
   opinion. Never present analyst multiplier tables as official X numbers; never invent
   metrics, engagement multiples, or "optimal" posting frequencies.
2. **No growth-hack garbage.** Never recommend: engagement pods/reply rings, follow-unfollow,
   mass auto-DMs, buying followers, rage-bait, copy-pasted spam replies, or anything designed
   to trigger negative signals (report/mute/not-interested).
3. **Replies are the highest-leverage, highest-risk move.** Each reply must be good enough to
   stand alone as its own post. If you wouldn't post it as a standalone, don't post it as a reply.
4. **Links last.** Any external link goes as a reply to the post or after the value is
   already delivered in the post body — never as the lead.
5. **Measure before theorizing.** No strategy recommendation without first reading the
   account's actual analytics.
6. **This skill gives strategy and analysis, never automation abuse.** It does not script
   bulk replies, scraping, or auto-engagement.