substack-skill

name	substack-skill
description	Create, edit, publish, schedule, and manage Substack posts. Use when the user asks to draft or publish a Substack post, list/edit their Substack drafts, schedule a post, or upload images to Substack. Supports multiple accounts/publications.
allowed-tools	Bash, Read

Substack Skill - Create, Edit, Publish, and Manage Posts

Manage Substack publications from the CLI via Substack's internal JSON API using your browser session cookie. Write posts in plain markdown — the skill converts to Substack's ProseMirror editor format (and back when fetching drafts).

CRITICAL: Publishing Confirmation Required

Before running publish or schedule, Claude MUST get explicit user confirmation:

1. Show the user the draft title + a summary of the body.
2. State clearly whether it will email all subscribers (--send-email) or publish to web only (--no-email). The email blast cannot be unsent.
3. Wait for an explicit yes.

Always confirm first. No exceptions. create-draft / update-draft are safe (drafts are private) and do not require confirmation.

First-Time Setup (One-Time)

1. Log in to Substack in your browser.
2. Open DevTools → Application → Cookies → https://substack.com and copy the value of the substack.sid cookie (long string starting with s%3A).
3. Run setup:

python3 ~/.claude/skills/substack-skill/substack_skill.py setup

It prompts for an account label, your publication URL (e.g. https://yourname.substack.com or your custom domain), and the cookie, then verifies auth and saves config.json. Sessions last ~3 months; when calls start returning auth_failed, grab a fresh cookie and re-run setup with the same label.

Multi-Account Convention

config.json (gitignored) mirrors notion-skill/zergscholar:

{
  "default_account": "personal",
  "accounts": {
    "personal": {
      "publication_url": "https://yourname.substack.com",
      "cookie": "substack.sid=s%3A...",
      "user_id": 123456,
      "user_name": "Your Name"
    },
    "company": { "publication_url": "https://blog.company.com", "cookie": "..." }
  }
}

Every command takes -a ACCOUNT to override the default.

Commands

Account

# Verify auth; show user, publications, subscriber count
python3 ~/.claude/skills/substack-skill/substack_skill.py whoami [-a ACCOUNT]

# List sections (for --section-id)
python3 ~/.claude/skills/substack-skill/substack_skill.py sections [-a ACCOUNT]

Reading

# List posts: --status published (default) | draft | scheduled
python3 ~/.claude/skills/substack-skill/substack_skill.py list-posts --status draft --limit 15

# Fetch a draft with its body converted back to markdown (--raw for full JSON)
python3 ~/.claude/skills/substack-skill/substack_skill.py get-draft DRAFT_ID

# Fetch a published post by slug
python3 ~/.claude/skills/substack-skill/substack_skill.py get-post my-post-slug

Writing

# Create a draft from a markdown file (local images auto-upload to Substack CDN)
python3 ~/.claude/skills/substack-skill/substack_skill.py create-draft \
  --title "Post Title" --subtitle "Optional subtitle" \
  --body-md ~/path/to/post.md [--audience everyone|only_paid|only_free|founding] [--section-id N]

# Or with inline markdown
python3 ~/.claude/skills/substack-skill/substack_skill.py create-draft \
  --title "Quick note" --body-text "Some **markdown** here."

# Update a draft (any subset of fields; body replace is full-overwrite)
python3 ~/.claude/skills/substack-skill/substack_skill.py update-draft DRAFT_ID \
  [--title ...] [--subtitle ...] [--body-md file.md] [--audience ...]

# Upload an image by itself, get the CDN URL
python3 ~/.claude/skills/substack-skill/substack_skill.py upload-image ~/pic.png

Publishing (confirm with user first)

# Publish AND email every subscriber (irreversible email blast)
python3 ~/.claude/skills/substack-skill/substack_skill.py publish DRAFT_ID --send-email [--share]

# Publish to web only, no email
python3 ~/.claude/skills/substack-skill/substack_skill.py publish DRAFT_ID --no-email

# Schedule / unschedule
python3 ~/.claude/skills/substack-skill/substack_skill.py schedule DRAFT_ID --at "2026-07-08T09:00:00-07:00"
python3 ~/.claude/skills/substack-skill/substack_skill.py unschedule DRAFT_ID

# Delete a draft (requires --force)
python3 ~/.claude/skills/substack-skill/substack_skill.py delete-draft DRAFT_ID --force

Examples

# Draft a post from a vault thought piece, review it, then publish with email
python3 ~/.claude/skills/substack-skill/substack_skill.py create-draft \
  --title "Nobody Reads Code Anymore" --body-md "$VAULT/Writing/nobody-reads-code.md"
python3 ~/.claude/skills/substack-skill/substack_skill.py get-draft 161803398
# ...user confirms...
python3 ~/.claude/skills/substack-skill/substack_skill.py publish 161803398 --send-email

Markdown Support

Headings, paragraphs, bold, italic, code, strike, links, bullet/ordered lists (one nesting level), blockquotes, fenced code blocks (with language), horizontal rules, and images. Standalone-line images become Substack image blocks (local file paths are uploaded automatically); inline images degrade to links. Not supported by Substack's format: tables, LaTeX — multi-paragraph list items are not supported by this converter.

Output

All commands print JSON to stdout and exit non-zero on error ({"error": code, "message": ...}; exit 2 = missing config).

Requirements

• Python 3.9+ (stdlib only — no pip installs)

API Limitations

• Unofficial API. Substack has no official post-management API; this uses the same internal endpoints as their web editor (as does the python-substack ecosystem). Endpoints can change without notice; the official "Developer API" is profile-lookup only and not a substitute.
• Session cookies last ~3 months, then re-run setup.
• update-draft --body-md replaces the whole body: editor-only widgets (buttons, polls, paywalls) in the existing draft are destroyed.
• No documented rate limits; the skill retries 429/5xx with backoff. Keep bulk operations gentle.

Security Notes

• Publishing confirmation required — Claude must always confirm with the user before publish or schedule (the --send-email blast is irreversible).
• The session cookie in config.json is equivalent to your logged-in browser session. It is stored locally and gitignored (blanket **/config.json rule).
• Revoke by logging out of that browser session or changing your password.