/skill

Connections Optimizer

Improve the user's network as a whole. Do not treat it as a cold lead list.

This skill can:

• Review X follows
• Review LinkedIn follows and connections
• Build a safe review queue
• Suggest people to follow or add
• Find warm paths to people
• Draft X DMs, LinkedIn notes, and Apple Mail messages
• Match each person to the best contact path

When to Use

Use this skill when the user:

• Wants to clean up X follows
• Wants to review LinkedIn connections
• Asks who to keep, unfollow, follow, add, or contact again
• Wants a network that fits their current goals
• Needs better warm outreach, not a cold lead list

Inputs

Get these facts before the review:

• The user's top goals
• The work the user is doing now
• Target jobs, fields, places, or groups
• Platform choice: X, LinkedIn, or both
• People or groups that must not be changed
• Mode: light-pass, default, or aggressive

Ask short questions when key facts are missing. Do not guess a person's value from weak data.

If the user does not pick a mode, use default.

Data and Tool Rules

Use only data and tools that the user has allowed.

Helpful tools may include:

• x-api for X follows and recent posts
• lead-intelligence for target ideas and warm paths
• social-graph-ranker for bridge value
• brand-voice before writing messages
• Browser tools when an API is not available
• Apple Mail or Mail.app for saved email drafts

If a tool is missing, work from data the user gives you. State what could not be checked.

Do not make up private data, shared contacts, recent activity, or past replies.

Safe Defaults

• Review comes before any change.
• Never send a DM, invite, or email on your own.
• Never follow, unfollow, remove, or block someone without clear approval.
• On X, review only accounts the user follows. Do not change followers.
• On LinkedIn, treat first-degree connections as high risk.
• Do not remove a first-degree connection by default.
• Show a ranked plan and all drafts before any action.
• Put all protected people on a Do Not Change list.
• Skip accounts when identity is not clear.
• Mark missing or old data as uncertain.
• Never use private traits to score people.

Platform Rules

X

• A mutual follow is often worth more than a one-way follow.
• A stale one-way follow may be a good review item.
• Show closed, missing, or long-idle accounts early.
• Value useful posts, real talks, and bridge value.
• Do not rank people by follower count alone.
• Do not treat a lack of follow-back as proof of low value.

LinkedIn

• Use the LinkedIn API only when the user has access and allows it.
• Use a browser review when API access is not available.
• Keep follows separate from first-degree connections.
• Follows can be cleaned up with less risk.
• First-degree connections must stay in the review queue unless the user approves removal.
• A job change, old title, or quiet profile is not enough reason to remove someone.

Modes

light-pass

• List only clear, low-value one-way follows for review.
• Keep uncertain cases out of the prune list.
• Suggest a small set of new people.

default

• Build a balanced prune queue.
• Build a balanced keep list.
• Rank people to add or follow.
• Draft warm outreach when it has a clear reason.

aggressive

• Build a larger prune queue.
• Give less weight to old one-way follows.
• Keep mutual ties and warm paths safer.
• Require review before every change.

Scoring

Give each person a score from 0 to 100.

Use these positive signs:

• Mutual connection
• Recent useful activity
• Fit with the user's current goals
• Link to another useful group
• Relevant job or field
• Real past talks or replies
• Signs that the person may respond

Use these negative signs:

• Closed or abandoned account
• Old one-way follow with no clear value
• Topics far from the user's goals
• Repeated low-value posts
• Many past messages with no reply
• Many stronger people who fill the same role

For each score, list the facts used. Lower the confidence when data is missing.

Do not give a strong penalty to:

• Mutual follows
• People tied to a real warm path
• Old friends or past coworkers
• Anyone on the Do Not Change list

Action Labels

Use one of these labels for each person:

• Protect: Never change without a new review
• Keep: Clear value
• Review: Mixed or missing facts
• Prune Candidate: Good reason to unfollow
• Add Candidate: Good reason to follow or connect
• Contact Candidate: Good reason to reach out now
• No Action: No useful next step

A score alone must not trigger an action.

Workflow

1. Get the user's goals, limits, platform, and mode.
2. Build a list of current follows and connections.
3. Remove duplicates and match profiles with care.
4. Set aside protected people and unclear profiles.
5. Score prune candidates. Give a short reason for each.
6. Score keep candidates. Give a short reason for each.
7. Rank people to add, follow, or contact.
8. Find any real warm path, such as a shared contact or past talk.
9. Pick the best contact path:
   • Use X DM for a short, warm social note.
   • Use LinkedIn for work ties and shared career links.
   • Use email for a longer note or a warm intro.
   • Use no message when contact would feel forced.
10. Use brand-voice if it is available. If not, copy the tone from user-approved samples.
11. Draft the messages. Do not send them.
12. Return the review pack.
13. Wait for clear approval before any change.

Edge Cases

• If the account is private, use only facts the user can see.
• If two profiles may be the same person, mark them for review.
• If a person changed jobs, score the person, not just the old title.
• If activity data is old, say when it was last checked.
• If there is no real warm path, write none found.
• If the user has no writing samples, use a plain and kind tone.
• If a draft needs a personal fact that is not known, add [add detail].
• If a person is useful but posts too much, suggest mute before unfollow.
• If a LinkedIn tie is weak but may matter later, suggest unfollow before removal.
• If the list is large, review it in small groups. Start with the clearest cases.
• If the user rejects a reason, update the rule before scoring the rest.

Review Pack Format

CONNECTIONS OPTIMIZER REPORT
============================

Mode:
Platforms:
Main Goals:
Data Checked:
Data Gaps:

Do Not Change
- profile:
  reason:

Prune Queue
- profile:
  score:
  reason:
  confidence:
  suggested action:

Review Queue
- profile:
  reason:
  missing facts:
  risk:

Keep / Protect
- profile:
  score:
  reason:
  bridge value:

Add / Follow Targets
- person:
  why now:
  warm path:
  best channel:
  confidence:

Drafts
- person:
  channel:
  draft:

Next Review Step
- action:
  needs approval:

Message Rules

• Save email as an Apple Mail or Mail.app draft when that option is allowed.
• Never send a message on your own.
• Keep the note short and clear.
• Say why the user is reaching out now.
• Use a real shared fact when one is known.
• Do not fake praise, closeness, or a warm path.
• Do not force a DM when no message is best.
• Make the draft sound like the user.
• Avoid sales phrases and fake urgency.
• Use one simple request at most.

Example

User request:

Clean up my X network in default mode. I now focus on small AI tools.
Do not change friends, past coworkers, or current customers.

Example result:

CONNECTIONS OPTIMIZER REPORT
============================

Mode: default
Platforms: X
Main Goals: Meet people who build small AI tools
Data Checked: User-provided follow list and recent post notes
Data Gaps: No DM history

Do Not Change
- profile: @sam
  reason: Past coworker

Prune Queue
- profile: @oldnewsfeed
  score: 24
  reason: One-way follow, no posts in 18 months, and no fit with the current goal
  confidence: high
  suggested action: Unfollow after review

Review Queue
- profile: @quietbuilder
  reason: Strong topic fit, but no recent activity data
  missing facts: Last post date
  risk: A useful person may be removed

Keep / Protect
- profile: @tinytools
  score: 88
  reason: Mutual follow, recent talks, and strong fit
  bridge value: Links the user to a small maker group

Add / Follow Targets
- person: Casey Lee
  why now: Builds small AI tools and knows two people in the user's network
  warm path: Shared contact, Jordan
  best channel: X DM
  confidence: medium

Drafts
- person: Casey Lee
  channel: X DM
  draft: Hi Casey, I have been working on small AI tools too. Jordan has shared good things about your work. I liked your post on simple setup. Want to compare notes sometime?

Next Review Step
- action: Approve or reject each prune item
  needs approval: yes

Related Skills

• brand-voice: Learn and reuse the user's writing style
• social-graph-ranker: Score bridge value and warm paths
• lead-intelligence: Find and rank people
• x-api: Review the X network and apply approved changes
• content-engine: Draft public posts about the user's network or launch plans