X account audit

Answers one question: what does this account post, how often, and what actually lands?

Most account "audits" average every item on the timeline. That number is wrong before anyone reads it, because a timeline is not a list of posts. It mixes the account's original posts with its replies, its retweets of other people, and the continuation tweets of its own threads. Each of those reaches a different audience and none of them is a post in the sense the person asking means.

This skill separates them first.

What it does

1. Confirms the handle back to you before spending anything. Read the profile with GET /twitter/user/info?username=<handle> and state the name, follower count and join date. Wait for a yes. A similar handle produces a confident audit of the wrong account.
2. Reads the recent timeline with GET /twitter/user/tweets?username=<handle>, following next_cursor until it covers the window you agreed (default: the last 30 days or 3 pages, whichever comes first). Say how many pages and items you read.
3. Classifies every item using the flags each one carries:
   • is_retweet: true is someone else's post. Report the count, never its metrics.
   • is_reply: true with in_reply_to_status_id pointing at the account's own tweet in the same conversation_id is a thread continuation.
   • is_reply: true to anyone else is a reply.
   • everything else is an original post. A quote tweet (is_quote: true) is the account's own post that embeds someone else's, so it counts as original; report how many originals were quotes.
4. Measures originals only, then reports.

The measures

Median, not mean. One post that travels moves the mean a long way and the median hardly at all. Report both and lead with the median; if they differ by more than 50 percent, say which post is pulling the mean and why.

The rule worth stealing even if you never install this

A thread continuation is not a post. An account that writes 1 post and 3 continuations has posted once. Counting the continuations as posts makes the account look four times as active and drags every median down, because continuations reach a fraction of the opener's audience.

That is measured, not hypothetical: see examples/real-run.md. On the run there, 3 of the first page's 17 items were the account's own continuations. Counting all 17 puts the median at 22,531 views; counting the 14 originals puts it at 46,072, about twice as high.

Hard boundary

Read-only. This skill never posts, likes, reposts, follows, sends a DM or drafts anything for you to send. It calls two GET endpoints and nothing else, even though the twitterapis API and MCP server can also write for an account you link. Write tools are not used by this skill and should not be enabled for it.

Output

The account line (name, handle, followers, join date), then the measures table, then the top posts, then one paragraph on what the top posts share (format, topic, media, time of day) that the median posts lack. End with what you could NOT conclude from public data (for example: impressions from ads, deleted posts, DMs).

Every number states its window and how many originals it rests on. A median of 4 posts is a weak claim; say so.

Setup

Needs a twitterapis.com API key in TWITTERAPIS_KEY, sent as the x-api-key header to https://api.twitterapis.com. Get one at https://www.twitterapis.com/signup; new accounts start with free credit and no card.

Calls per run, all reads at $0.0008 each: 1 profile read plus 1 per timeline page. A 3-page audit is 4 calls, about a third of a cent. Say the call count in the report.