Sylph Setup Skill

Setup Skill

MCP connectors

Core principle: ask less, derive more

Ask the user as few things as possible. For most channels, the agent can find existing content online (website, LinkedIn, blog, Twitter, Substack). Pull it, analyze it, derive insights and voice - then show the user for confirmation. Only ask when you genuinely can't find something.

When to use

Run /sylph-setup-skill <name> after completing /sylph-setup (global setup). Run it once per channel or skill you want to activate.

Usage

Copy code
/sylph-setup-skill linkedin
/sylph-setup-skill blog
/sylph-setup-skill newsletter
/sylph-setup-skill outbound
/sylph-setup-skill crm
...

How it works

1. Read the setup guide for the requested skill (from this folder)
2. Auto-discover and pull existing content from the web or connected repos
3. Copy the best examples to _published/ and _examples/
4. Derive insights (voice, tone, format, what works) from the examples
5. Update the skill's SKILL.md with derived rules
6. Update the channel's _insights.md with derived patterns
7. Show the user for confirmation and corrections

Available setup guides

Each guide is a markdown file in this folder:

Dispatch logic

When the user runs /sylph-setup-skill <name>:

1. Map the name to the corresponding setup guide file:

   • linkedin -> read setup-linkedin.md
   • blog -> read setup-blog.md
   • newsletter -> read setup-newsletter.md
   • substack -> read setup-substack.md
   • x or twitter -> read setup-x.md
   • reddit -> read setup-reddit.md
   • website -> read setup-website.md
   • email -> read setup-email.md
   • outbound -> read setup-outbound.md
   • crm -> read setup-crm.md
   • events -> read setup-events.md
   • investor-update or finance -> read setup-investor-update.md
   • brand -> read setup-brand.md
   • hr or screening -> read setup-hr.md
   • slack-community or slack -> read setup-slack-community.md
   • customer-report -> read setup-customer-report.md
   • customer-success or cs -> read setup-customer-success.md
   • zero-inbox or inbox -> read setup-zero-inbox.md
   • create-issue or issues -> read setup-create-issue.md
   • contracts or create-contract or review-contract -> read setup-contracts.md

2. Follow the guide step by step

3. After setup, generate a test piece to verify quality

If the name doesn't match any guide, list available options and ask the user to pick one.

Universal setup pattern

Every skill setup follows the same 4-step pattern:

Step 1: Auto-discover existing content

Before asking the user anything, search for and pull her existing content:

• LinkedIn: search her company LinkedIn page, read recent posts
• Blog: ask for the repo URL or blog URL, clone/read articles, copy to _published/
• Twitter/X: search her company/personal handle, read recent tweets
• Substack: find her Substack, read recent articles
• Newsletter: ask for archive URL or pull from Substack/website
• Website: read key pages from the company website (already in CONTEXT.md)

For each piece found:

• Save to <channel>/_published/YYYY-MM-DD_slug.md with proper frontmatter
• Promote the 2-3 best to <channel>/_examples/

Step 2: Derive insights automatically

From the collected content, analyze and write to <channel>/_insights.md:

• Voice patterns: sentence length, vocabulary, tone, favorite constructions
• Format patterns: what types of content she writes most, how she structures them
• Topic patterns: recurring themes, angles, positioning
• What works: engagement signals (likes, comments, shares if visible)
• Cadence: how often she publishes (derive from dates)

Show the insights to the user:

Here's what I learned from your existing [channel] content. What would you change?

Step 3: Configure the skill

Update the skill's SKILL.md with derived rules:

• Voice rules specific to this channel (from analyzed content)
• Format constraints (from patterns observed)
• Platform-specific rules (from what she actually does)

Step 4: Test generation

Generate one piece of content using the newly configured skill. Ask the user to review. Capture any edits as the first learning cycle in _insights.md.