linkedin-posting

name	linkedin-posting
description	Use when the user wants to draft and publish a post on LinkedIn. Interviews them, drafts in their voice, and publishes on approval. Handles first-time LinkedIn app + OAuth setup automatically.

LinkedIn Posting

Draft and publish LinkedIn posts from inside Claude Code. This skill walks a user from "I want to post about X" to a published post on their feed, including first-time authentication.

When to use

Trigger on any of: "post this to LinkedIn", "draft a LinkedIn post", "share on LinkedIn", "post on my feed about …". Not for fetching posts or analytics — this skill only publishes.

Flow

┌──────────────────────────────────────────┐
│ 1. Check auth status                     │
│    python scripts/linkai_auth.py status  │
└──────────────────────────────────────────┘
                   │
     ┌─────────────┼─────────────┬──────────────────┐
     ▼             ▼             ▼                  ▼
  no creds     missing token  expired token      valid
     │             │             │                  │
     ▼             ▼             ▼                  │
 ONBOARDING      LOGIN         LOGIN                │
 (below)      (below)        (below)                │
     └─────────────┴─────────────┴──────────────────┘
                                                    ▼
                                          ┌──────────────────┐
                                          │ 2. Interview     │
                                          │ 3. Draft         │
                                          │ 4. Revise        │
                                          │ 5. Publish       │
                                          └──────────────────┘

1. Check auth status first (this is the onboarding gate)

Before anything else, run the status check. The scripts live in scripts/ relative to this SKILL.md — invoke them with their absolute path via Bash.

python "<skill-dir>/scripts/linkai_auth.py" status

The output tells you exactly which branch to take. Do not start onboarding unless the output says "No credentials found". Running onboarding on a user who has already set things up wastes their time.

status output	What it means	What to do
No credentials found at …	First-time user	Full onboarding (section below)
App credentials: present + Access token: not yet issued	Dev app exists, never logged in	Skip to Step B: LinkedIn login
Access token: EXPIRED	Token issued >60 days ago	Skip to Step B: LinkedIn login
Access token: valid (N days remaining)	All good	Skip to the posting interview

Onboarding (only when status says "No credentials found")

Browser-driven and paced. Do not dump every step on the user at once. Open the relevant URL, wait for their confirmation, move to the next step. Use this Bash one-liner to open a URL in their browser on any OS:

python -c "import webbrowser; webbrowser.open('<url>')"

Before the walkthrough, ask one question: "Do you already have a LinkedIn developer app you'd like to reuse, or should we create a new one together?"

• If reuse: they need Client ID + Client Secret from their existing app, and http://localhost:8765 in its Authorized redirect URLs. Skip to Step A: Collect credentials, but first remind them to add the redirect URI if it's not already there and open their app list with:

  python -c "import webbrowser; webbrowser.open('https://www.linkedin.com/developers/apps')"
  

• If create new: proceed with the full walkthrough below.

Create app (only if they chose "create new")

Step 1 — open the new-app page.

python -c "import webbrowser; webbrowser.open('https://www.linkedin.com/developers/apps/new')"

Tell them: "I just opened the LinkedIn Create app form. Fill in App name (anything, e.g. 'LinkAI for '), pick or create a LinkedIn Page (if you don't have one, click the inline Create a new LinkedIn Page link — a bare personal-brand page is fine; the app uses it for admin only, posts still go to your personal feed), add any Privacy policy URL (e.g. this repo's README link), upload any small image for App logo, then click Create app. Tell me when you're on the app's detail page."

Wait for confirmation. If they hit a verification prompt or get stuck, help them through it — don't move on.

Step 2 — attach the two products. Once they're in the app, tell them to open the Products tab. Ask them to confirm they're on it, then:

• "Find Sign In with LinkedIn using OpenID Connect and click Request access. It's instant."
• "Then find Share on LinkedIn and click Request access. Also instant."

Wait until both show "Added".

Step 3 — add the redirect URI. Tell them to open the Auth tab. Under OAuth 2.0 settings → Authorized redirect URLs for your app, they click the edit pencil and add exactly:

http://localhost:8765

No trailing slash, no www, no https. Save.

Step 4 — grab Client ID and Client Secret. Still on the Auth tab: tell them the Client ID and Primary Client Secret are right at the top of the page. Ask them to paste both into the chat.

Step A: Collect credentials

Once you have Client ID and Client Secret (from either the reuse or create path), run:

python "<skill-dir>/scripts/linkai_auth.py" setup --client-id "<id>" --client-secret "<secret>"

Expect: Saved app credentials to ~/.linkai/credentials.json.

Step B: LinkedIn login

python "<skill-dir>/scripts/linkai_auth.py" login

The browser opens, the user clicks Allow, the browser lands on a "LinkAI is authorized" page, and the CLI prints Logged in as <Name>. Relay the CLI output if anything goes wrong — don't silently retry.

When this succeeds, transition to the posting interview.

Normal posting flow

When auth is good, run the four steps in order. Do not short-circuit.

1. Interview. Ask targeted questions to get the post's substance: what happened / what they're sharing, why it matters, who it's for, any specifics worth naming (people, companies, numbers, links). Probe for concrete details. Keep each question short and ask them one at a time unless clearly bundled.
2. Draft. Write the post in the user's voice based on what they told you. Match the tone they used in the interview — energetic if they were energetic, measured if they were measured. No "As an AI" framing, no forced persona, no opening "Hi, I'm …". Just a post that reads like the user wrote it.
3. Revise. Show the draft and ask for feedback. Iterate until they approve. Don't push back on their edits unless you have a concrete reason.
4. Publish. Only after explicit approval (e.g., "yes post it", "looks good, send it"). Invoke:

python "<skill-dir>/scripts/linkai_post.py" --text "<final post text>"

The CLI prints a URL like https://www.linkedin.com/feed/update/urn:li:share:…/. Share it with the user.

Voice and tone

• Write as the user, not as a character. No personas, no signatures like "— Claude" or "— Link".
• Default to professional-but-human. Not corny, not corporate-hollow, not cliché-heavy.
• Match length to content: a quick thought is two sentences; a real story is four short paragraphs. Don't pad.
• Use the user's own phrases when they said something well during the interview.
• Emoji only if the user's inputs suggest they use them.

Error recovery

• HTTP 401 from post CLI → it auto-runs login once and retries. No action needed.
• Post CLI exits non-zero with "Credentials incomplete" → fall back to onboarding.
• Login fails with "Could not bind 127.0.0.1:8765" → another process (or a previous hung login) is holding the port. Ask the user to close other LinkAI auth windows or restart their terminal, then retry.
• Login fails with "No callback received within 300s" → the user didn't complete the browser flow in time. Offer to retry.
• LinkedIn returns 403 on post → scope or product missing on the dev app. Point to SETUP.md and verify Share on LinkedIn is attached.

Notes

• Tokens are 60-day; re-auth is a single browser click and the skill does it automatically on 401 — you don't need to warn the user unless they ask why the browser opened.
• Credentials live at ~/.linkai/credentials.json (permissioned 0600 on Unix). Never print the file's contents in chat.
• Text-only posts only. If the user wants to attach an image, article, or poll, tell them that's not supported in this version and offer to post text-only instead.