List Pruner

Runs on: reasoning - reads your files and reasons; any capable agent can run this.

Prepare a contact list for outreach planning. The skill removes duplicate emails, flags missing fields, assigns a quality score, and returns a clean markdown table or CSV-ready output.

Pre-flight

If core/identity.md does not exist, stop with: Founder OS not set up here. Run /founder-os:setup first.

When to invoke

Invoke when you provide a CSV, a pasted table, or a path to a contact list before outreach. Also invoke during a list review when a campaign list may contain stale rows.

Protocol

1. Read the input list from a CSV path or a pasted table.
2. Confirm the required columns are present, or map similar columns to name, email, company, title, persona, and signal.
3. Normalize emails to lowercase.
4. Remove duplicate rows by email. Keep the row with the most filled fields.
5. Flag any missing email, company, title, persona, or signal.
6. Score each row:
   • High when email, company, title, persona, and signal are all present.
   • Medium when email is present and exactly one non-email field is missing.
   • Low when email is missing, or two or more non-email fields are missing.
7. Return a clean table with score and flag columns.
8. Write a clean CSV only if you ask for a file. If writing, confirm the output path first.
9. Append a one-line trace to brain/log.md (#acted tag) recording input rows, clean rows, and duplicates removed.

Output schema

list_pruner:
  status: ok | blocked
  input_rows: <number>
  clean_rows: <number>
  duplicates_removed: <number>
  flagged_rows: <number>
  output_table:
    - name: "<name>"
      email: "<email@example.com>"
      company: "<company>"
      title: "<title>"
      persona: "<persona>"
      signal: "<signal>"
      score: High | Medium | Low
      flag: "<none | missing_email | missing_company | missing_title | missing_persona | missing_signal>"

Failure modes

• No email column: proceed only if you confirm the outreach channel is not email.
• Unreadable CSV: stop and ask for a pasted table or a valid path.
• Empty list: stop with blocked_empty_list.
• Real credentials found in the file: stop and report the path. Do not echo the secret.
• File write requested but output path missing: ask for a path before writing.

Composes with

• linkedin-network-scan, which produces a ranked worklist from your LinkedIn export - this skill cleans a list assembled by hand or from another source before you act on it.
• email-drafter, which drafts outreach to the cleaned rows one at a time.
• context/leads.md, the OS leads file. After pruning, the High-scored rows are the natural candidates to add as new Stage: Raw leads. The skill does not auto-write that file; it surfaces the candidates and you add them through your normal leads flow.

Free-tier path

Free-tier users paste the CSV contents straight into chat. The model returns the clean table for manual copy into a spreadsheet or into context/leads.md. File writes are optional, not required.

Worked example

Input:

name,email,company,title,persona,signal
Jordan Lee,jordan@example.com,Example Labs,Head of Ops,operator,hiring ops roles
Jordan Lee,JORDAN@example.com,Example Labs,,operator,hiring ops roles
Mira Patel,mira@example.com,Northstar Demo,Founder,builder,posted about outbound

Output:

list_pruner:
  status: ok
  input_rows: 3
  clean_rows: 2
  duplicates_removed: 1
  flagged_rows: 0
  output_table:
    - name: Jordan Lee
      email: jordan@example.com
      company: Example Labs
      title: Head of Ops
      persona: operator
      signal: hiring ops roles
      score: High
      flag: none
    - name: Mira Patel
      email: mira@example.com
      company: Northstar Demo
      title: Founder
      persona: builder
      signal: posted about outbound
      score: High
      flag: none

Cross-reference

• skills/linkedin-network-scan/SKILL.md - builds a ranked list from your own network export.
• skills/email-drafter/SKILL.md - drafts the outreach to each clean row.
• context/leads.md - where High-scored rows become tracked leads.