The Shelf / Growth / X Growth System
X Growth System
Run the X API two ways: read the firehose for pennies, post under cost + ban governors.
The job: turn X's pay-per-use API into two disciplined tools — a research firehose that costs pennies, and a growth channel that doesn't get your account suspended or your credit balance drained.
X quietly became the best demand-research corpus available to a solo builder: full-archive search over people saying exactly what they wish existed. It also became a meter — every API call bills, and a single post containing a link costs about 13x a plain post. Most people discover both facts the expensive way. This skill front-loads them.
What it makes your agent do
- Lead with research. Before any posting, the agent mines the firehose: verbatim demand quotes ("is there an app that..."), willingness-to-pay signals, competitor complaints that become roadmap items, and the discovery-path verdict — does content in this niche actually get engagement, or is there no audience to grow into? A full dossier runs on cents.
- State cost before spending. Every run gets a pre-flight estimate from the included price table. Over the cap? The agent stops and asks. After the run, estimate vs. actual gets logged so the estimator stays honest.
- Ration the expensive call. The link-in-pinned tactic alone pays for the skill: one paid link post, pinned, with cheap text posts pointing at it — versus paying link price on every promo post.
- Queue writes, never blast them. Each draft carries the exact text, its cost, and a policy checklist mapped to X's real automation rules — no duplicate copy across accounts, no keyword-triggered replies, no mass-tagging, no DMs. Fail a box, it doesn't queue. You approve per item; dry-run is the default.
- Skip the setup minefield. The included playbook for the official hosted X MCP documents eight specific failures from a real setup — the app-not-in-a-Project trap, the read-only default, the port 8080 collision, the two-account credit trap — each with its fix.
Who it's for
Builders who want evidence before code. Indie devs launching on X with one brand account and a small credit balance. Agent operators who want posting power with the safety catch on.
Why it beats a free directory download
Free X-marketing prompts don't know the API bills per call, don't know which automation patterns suspend sanctioned accounts, and haven't walked the MCP setup failures. This system encodes all three from real operation, and it's maintained as X's pricing and rules move. The field report below shows the cost math working before you pay.
FIELD REPORT real output, not a promise
The cost governor doing its job: real pricing pulled from X's published pay-per-use table (verified against docs.x.com, mid-2026) and the launch-thread math the skill runs before a single post goes out.
The price table the agent works from
| Call | Price |
|---|---|
| Owned read (your own posts, followers, bookmarks) | $0.001 |
| Post read (others' posts, search results) | $0.005 |
| User-tied read (profiles, trends) | $0.010 |
| Write, no URL | $0.015 |
| Write containing a URL | $0.20 |
One number dominates: a link-post costs 200x an owned read and 13x a plain post.
Pre-flight on a 6-post launch thread
Naive plan — link in every post of the thread:
6 posts x $0.20 (each contains the store URL) = $1.20
Governed plan — the skill's link-in-pinned tactic:
1 pinned post with the URL = $0.20 (paid once)
5 thread posts, plain text,
pointing at "link in pinned/bio" = 5 x $0.015 = $0.075
Total = $0.275
Same launch, same link, 77% less spend — and the pinned link keeps paying: every future promo post references it at plain-text price instead of re-buying the URL.
Pre-flight on a research dossier
Demand mining: 10 full-archive searches x $0.005 = $0.05
Discovery-path scan: ~15 reads incl. user-tied ≈ $0.10
Competitor pass: ~20 reads ≈ $0.10-0.20
An entire niche dossier — demand quotes, discovery-path verdict, competitor signal — for roughly the price of two plain posts. That asymmetry is the skill's whole thesis: read heavily, write rarely, and never let a loop touch the $0.20 call.
The checklist every write clears before queueing
Not duplicative across accounts · not a keyword-triggered auto-reply · not a mass-tag · not a DM · disclosed if promo · posted from the confirmed designated account · cost approved. Fail one box, the item doesn't queue.
SERVICE RECORD living gear — updated as the factory learns
v1.0.0 — 2026-07-17
First issue. Ported from the factory's internal skill: sanitized for general use, methodology intact, field report captured from a real run.
Every update ships free to owners — your locker always serves the latest version.
QUESTIONS
Do I need to pay X to use this?›
For the API path, yes — X's API is pay-per-use now, no free tier. The point of the skill is that reads cost fractions of a cent, so research is nearly free, and the cost governor keeps writes rationed. A few dollars of credits proves the whole pipe.
Will the agent post on its own?›
No. Dry-run is the default. Every write is drafted into a queue with its cost and a policy checklist; you approve per item before anything goes out.
Is API posting allowed by X?›
Yes — posting your own content via approved API access is sanctioned; the skill uses X's official hosted MCP. What gets accounts suspended is spam patterns (duplicate copy, auto-replies, mass-tagging, auto-DMs), and the governor blocks exactly those.
Which platforms does it run on?›
The MCP wiring targets Claude Code or any MCP-capable agent. The methodology files (governors, ladder, templates) are plain markdown and work in claude.ai Projects or Codex too — you just execute the posts yourself there.