post-to-linkedin

name	post-to-linkedin
description	Use when announcing a release, paper, grant, talk, award, or project update on LinkedIn from this repo. Covers drafting, OAuth setup, token refresh, and posting via scripts/post_to_linkedin.py.

Post to LinkedIn

Compose and post announcements to LinkedIn from this repo via scripts/post_to_linkedin.py. LinkedIn API posts are free (no per-post charge, unlike X Premium). Max commentary length is 3,000 chars.

The draft format, hashtag sets, and writing rules are shared with the post-to-x skill (same scripts/drafts/ directory, same references/draft-patterns.md, same references/hashtag-sets.md at skills/post-to-x/references/).

When to use

• User asks to post, share, or announce something on LinkedIn
• Cross-post: the same release/paper/grant that went to X, adapted for LinkedIn tone
• Longer-form academic/professional framing where the 25,000-char X Premium post would feel out of place

Do not use for: generic networking DMs, connection requests, group posts to a Company Page feed (this skill posts to the authenticated user's personal feed only).

Prerequisites

Two gates.

Gate 1: drafting / previewing

• ~/miniforge3/envs/py312/python.exe has requests and python-dotenv installed (pip install -r scripts/requirements-post.txt).
• Source data available: data/open-source.json, data/publications.json, or user-provided text.

Gate 2: real post

• .env at repo root contains LINKEDIN_CLIENT_ID + LINKEDIN_CLIENT_SECRET.
• LinkedIn app has products Share on LinkedIn (Default Tier, gives w_member_social) and Sign In with LinkedIn using OpenID Connect (Standard Tier, gives openid profile email) approved. Both are self-serve; approval is typically instant.
• .env also contains LINKEDIN_ACCESS_TOKEN, LINKEDIN_TOKEN_EXPIRES_AT, LINKEDIN_USER_URN, written by --auth on first run. LINKEDIN_REFRESH_TOKEN is written when LinkedIn returns one; on default self-serve apps it may be empty, in which case re-run --auth after the access token expires (60 days).
• User has reviewed and approved the dry-run preview for this specific post.

If Gate 2 fails, stop and report before posting. Drafting and previewing may continue.

One-time OAuth setup

Before the first post, user must:

1. Create a LinkedIn Developer app at https://developer.linkedin.com (requires a Company Page, not a personal profile; a 2-minute free Company Page creation is fine).
2. Add Products: Share on LinkedIn + Sign In with LinkedIn using OpenID Connect. Both auto-approve.
3. In the app's Auth tab, add http://localhost:8765/callback to Authorized redirect URLs.
4. Copy Client ID + Client Secret into .env (see .env.example).
5. Run:

   python scripts/post_to_linkedin.py --auth
   

   This opens a browser, user authorizes, tokens + user URN are written back to .env. Takes about 10 seconds.

Access tokens last 60 days. The script auto-refreshes when a refresh token is available, but LinkedIn's default self-serve tier typically does not return a refresh token (observed 2026-04-22). In that case, re-run --auth when the access token expires; the re-auth flow is unchanged, takes ~10 seconds.

Workflow

1. Identify source   →  open-source.json / publications.json / user-provided
2. Draft to file     →  scripts/drafts/<slug>.md  (same directory as X drafts)
                        Can reuse an X draft directly, or tweak for LinkedIn tone.
                        For LinkedIn the draft can be longer (up to 3,000 chars)
                        and more narrative; hashtag volume can be higher (~5 to 10
                        is normal on LinkedIn without penalty).
3. Preview dry-run   →  python scripts/post_to_linkedin.py --dry-run --draft <path>
4. Iterate           →  show preview, accept user edits, re-preview
5. Post              →  python scripts/post_to_linkedin.py --yes --draft <path>
                        Attach images with --media <path> (repeatable, up to 20).
6. Report URL        →  return https://www.linkedin.com/feed/update/<URN>/

Always run step 3 before step 5. No exceptions. --yes skips the interactive prompt; never use --yes without a prior successful --dry-run that the user approved.

Hard rules

• Lead with the one-line value prop, same as X posts. LinkedIn readers scan the first sentence; if it does not answer "what is different now?", they scroll past.
• LinkedIn tone is slightly more professional than X. Fewer bullet lists, more connective prose. Emojis are accepted but not required.
• Hashtags: 3 to 10 tags is fine on LinkedIn (higher ceiling than X). Pick from skills/post-to-x/references/hashtag-sets.md.
• Apply CLAUDE.md writing rules:
  • No em dashes (Unicode U+2014) or en dashes (Unicode U+2013) as casual punctuation.
  • No banned AI-tell words.
  • Prefer full forms (it is) over contractions.
  • No Unicode U+202F.
• Verify every factual claim. Star counts, download counts, venue names, coauthor names must match data/open-source.json, data/publications.json, or user-stated facts.
• Cost: zero. LinkedIn does not charge per-post via the API.

Post type differences from X

Aspect	X (Premium)	LinkedIn
Char limit	25,000	3,000
Per-post cost	$0.20 with URL, $0.015 without	free
URL length counting	t.co shortens to 23 chars	full literal length
Hashtags	3 to 5	3 to 10
Tone	direct, punchy, dev-adjacent	professional, narrative, network-adjacent
Thread support	native reply-chain (X side only)	not used; single post is canonical on LinkedIn
Image limit per post	4	20

Common mistakes (draft from post-to-x + observed)

Mistake	Fix
Reusing an X draft verbatim without adjusting tone	LinkedIn readers expect more context; add 1-2 sentences of positioning before the bullets
Dumping too many hashtags (over 10)	LinkedIn allows more than X but the algorithm still penalizes stuffing; stay at 5 to 8
Posting without running --dry-run	Contract: dry-run is mandatory. Check cost, char count, media attachments
Posting with expired token + no refresh_token	Re-run --auth to get a fresh refresh_token
Posting an X draft that uses t.co-shortened URL math	LinkedIn counts URLs literally; re-check char count on the preview

Quick reference

• Script source: scripts/post_to_linkedin.py
• Flags: python scripts/post_to_linkedin.py --help
• Env vars: see .env.example for LINKEDIN_* keys
• Shared draft patterns: skills/post-to-x/references/draft-patterns.md
• Shared hashtag sets: skills/post-to-x/references/hashtag-sets.md
• First live post: https://www.linkedin.com/feed/update/urn:li:share:7472734205077053440/ (posted 2026-06-16; "Safe Models, Unsafe Agents" talk announcement, slide image + 6-paper list)