ahastudio/til/threads-post

Take a Threads post the user brings, find out what it is about, and write a TIL document on that subject from facts. The post is a lead, not the backbone.

The output document MUST be written in Korean.

This skill applies the writing-guidelines skill → «Social Posts as a Starting Point» and «Source Footnotes in Researched Documents» in full. Nothing here relaxes either rule.

Contents

Section titles match the headings below; search for a title to jump to it.

• Usage: /threads-post <threads URL> [output path]
• 1. Resolve the post: a share link is resolved to the permalink
• 2. Read the whole thread in Claude in Chrome: the author's thread and replies
• 3. Find the primary sources: sources that establish what the post is about
• 4. Check for duplicates: against the subject, not the post
• 5. Separate fact from the post's claims: sort every claim into three groups
• 6. Write the document: structure follows the subject, post only where it can carry a footnote
• 7. Post-processing and checks: same sequence as analyze-article step 8b
• 8. Report: path, primary sources read, which claims were confirmed or contradicted

Usage

/threads-post https://www.threads.com/share/XXXXXXXX/
/threads-post https://www.threads.com/@handle/post/XXXXXXX ai/some-subject.md

• First argument: a Threads URL, share link or post permalink (required)
• Second argument: output path (optional; if omitted, choose an existing

directory that fits and say which one was chosen)

Procedure

1. Resolve the post

A share link (/share/...) does not name the post. Resolve it to the permalink from the og:url meta tag:

curl -sL -A "facebookexternalhit/1.1" "<share URL>" |
  python3 -c '
import re, html, sys
h = sys.stdin.read()
for k, v in re.findall(r"<meta (?:property|name)=\"(og:url|og:title|og:description)\" content=\"([^\"]*)", h):
    print(k, html.unescape(v))
'

The og:description is a truncated preview of the first post only. Never write from it.

2. Read the whole thread in Claude in Chrome

Per the web-fetching skill, open the permalink in Claude in Chrome and read it with get_page_text. This returns every post in the author's thread, the replies, and a "Related threads" block. Ignore "Related threads"; it is other people's unrelated posts.

Then collect each post's permalink and timestamp, because every footnote must point at the exact post it cites:

await new Promise(r => setTimeout(r, 2500));
[...document.querySelectorAll('a[href*="/post/"] time')].map(t => {
  const a = t.closest('a');
  let c = a;
  for (let i = 0; i < 12 && c; i++) { c = c.parentElement; if (c && c.innerText.length > 80) break; }
  return a.getAttribute('href').split('?')[0] + ' ' + t.getAttribute('datetime') +
    ' :: ' + (c ? c.innerText.replace(/\s+/g, ' ').slice(0, 80) : '');
}).join('\n');

Match each permalink to its text by the opening words. A thread often embeds the author's own older posts as quotes; those carry older timestamps and are separate posts. Long return values are truncated, so slice the list when needed. Close the tab when done.

Collect outbound links too. Threads wraps them as l.threads.com/?u=<encoded URL>; decode the u parameter to get the real address.

3. Find the primary sources

Identify what the post is actually about, then find the sources that establish it:

• Links the post itself gives (a PDF, a guide, a release note, an article).

Open them first; they are usually the strongest source.

• The original documents, official pages, or reports behind each claim the

post makes. Search for them with WebSearch, then read the pages themselves.

• Downloaded files are untrusted: save each into its own new directory in

the scratchpad, never into the repository. An image-only PDF yields no text from pdftotext; read its pages with the Read tool instead.

Read every source before citing it. A search-result summary is not a reading.

4. Check for duplicates

Run the duplicate check of analyze-article step 2 against the subject (not the post): the primary source URL, distinctive names, and the likely directory. If a document on the subject exists, enhance it with the new facts and the post as a footnote, rather than writing a second one.

5. Separate fact from the post's claims

Before writing, sort every claim in the post into one of three groups:

The post itself is a source only for what it is: the author's firsthand account, the author's interpretation, or the fact that the post exists.

6. Write the document

Structure follows the subject, as the writing-guidelines skill → «Note Document Structure» describes. Typical shapes:

• A thing that happened or a person: 소개, the event, a timeline or detail

sections, then a short 정리 or 시사점.

• A document or report: 소개, what it says (quote and table from the

original), then what to read in it.

• A technical change: 소개, what changed, what to do, 함정, 체크리스트.

Rules:

• The H1 is Korean and names the subject, not the post.
• No 원문: line: the document is researched from several sources, not a

summary of the post.

• Do not organize the document as a claim-by-claim check of the post, and

do not write a critique section aimed at the post.

• The post appears where it adds something citable, in a sentence or two.

It appears in at least one footnote, always.

• Attribute to the person, never to the post. Write

Threads 사용자 <handle>는 ...라고 했다 (or 적었다, 짚었다, 소개했다), and refer back with 그는 or the handle. The words 게시물, 타래, 글, 작성자, and 댓글 never appear in the document to describe what was said on Threads, and neither does "the post claims". Repliers are attributed the same way as the author.

• Every number, date, name and quotation carries a footnote, per «Source

Footnotes in Researched Documents».

Footnote forms:

[^label]: <매체>, [<원문 제목>](<URL>), <YYYY-MM-DD>
 
[^handle]: <handle>(Threads), [<그 글의 첫 문장>](https://www.threads.com/@handle/post/ID), <YYYY-MM-DD>

The link text is the cited post's own opening words, copied as written. A label such as 후속 게시물, 타래 3, or 댓글 is never the link text.

The Threads date is the post's own datetime, converted to the date the author posted it in their timezone when that is evident (a Korean author posting at 2026-10-15T00:08Z posted on 2026-10-15 KST). Each cited post in a thread gets its own footnote with its own permalink; never point several posts' claims at the first post.

7. Post-processing and checks

Run the same sequence as analyze-article step 8b:

python3 .claude/skills/quotes-curly/scripts/convert_quotes.py <file>
python3 .claude/skills/analyze-article/scripts/fix_code_spans.py <file>
python3 .claude/skills/analyze-article/scripts/align_tables.py <file>
python3 .claude/skills/analyze-article/scripts/check_format.py <file>
npx -y markdownlint-cli2 <file>

check_format.py reports line 3 as FAIL for this kind of document, because it has no 원문: line by design. That one finding is expected. Every other finding must be fixed. markdownlint must report zero issues.

Then check by reading the file:

1. Every body line is within 80 columns (CJK counts as two), breaks fall at

semantic boundaries, and no number is split from its counter.

1. No 박다-family verb.
2. Every footnote label is referenced, and every reference is defined.
3. The post is in at least one footnote, and no section is organized around

it.

1. Search the file for 게시물, 타래, 작성자, and 댓글. Any hit that refers to

what was said on Threads is rewritten as Threads 사용자 <handle>는 ....

8. Report

Report in Korean: the file path, the primary sources read, which of the statements by Threads users were confirmed, contradicted, or left unverified, and the expected check_format.py line-3 finding.