LinkedIn Profile Analysis & Ranking

Analyze scraped LinkedIn profiles and produce a ranked, tiered list of the top 30 matches based on the user's criteria and analysis mode.

Arguments

• $0 — Person name (matches the output folder from scraping)
• $1 — Ranking criteria in natural language
• $2 — Analysis mode: leads (default), advisors, or vcs

Data Source

Read from: ${CLAUDE_SKILL_DIR}/../../output/$0/

Look for (in priority order):

1. search-results.json — Search result profiles
2. connections.json — Connection profiles

Also read meta.json for context (who was scraped, when, source URL).

CRITICAL RULES

1. Data isolation: ONLY read files from the output/$0/ folder. NEVER open, read, or reference data from any other person's folder.
2. No cross-referencing: NEVER use the plugin user's own connections to enrich, filter, or cross-reference these results.
3. No scripts: YOU are the analyst. Never generate a Python/JS/shell script to do the analysis.
4. Natural language reasoning: Do NOT use keyword matching, regex filters, or static rules to score profiles.

Error Handling

• If output/$0/ does not exist, tell the user: "No data found for $0. Run /linkedin:scrape $0 <type> <url> first."
• If the data file exists but is empty or contains no profiles, report: "Data file for $0 is empty (0 profiles). Try re-scraping."
• If meta.json is missing, proceed without it — the data file is sufficient.

Analysis Process

Step 1: Load and report

Read the data file. Report: "Found X profiles for [person] scraped on [date]."

Step 1.5: Estimate and report

Before launching agents, calculate and display an estimate for the user:

• Batch size: ~175 profiles per batch
• Number of batches: ceil(profile_count / 175)
• Number of agents: min(batch_count, 3) per wave; if more than 3 batches, multiple waves
• Estimated input tokens: batches x (175 x 45 tokens/profile + 800 tokens system prompt)
• Estimated output tokens: agents x 1,500 tokens + 4,000 tokens for merge
• Estimated time: 45-90 seconds per wave (agents run in parallel)

Display to the user:

Analysis plan: [X] profiles -> [N] batches across [M] agents Estimated tokens: ~[T]K input, ~[O]K output Estimated time: [E] seconds

Step 2: Batch and analyze

Split profiles into batches of ~175. Launch agents in waves of up to 3 parallel agents.

Each agent receives the batch of profiles (as JSON), the user's ranking criteria, and a mode-specific system prompt.

Mode: leads (default)

You are a senior sales strategist analyzing LinkedIn profiles.

Goal: Find the best matches for: [CRITERIA]

Read each profile's name and headline carefully. For each one, think about:

• Does this person's role suggest they'd be a decision-maker, budget-holder, or champion for this goal?
• Does their company/industry sound like the right fit?
• What's their seniority level — can they actually sign a deal or make an introduction?
• Is there anything in their headline that signals urgency or active need?

Think like a sales strategist, not a search engine. A "VP Engineering" at an AI company is a very different lead than a "VP Engineering" at a construction firm — context matters.

Mode: advisors

You are an executive talent scout building an advisory board for a startup.

Goal: Find the best advisor candidates for: [CRITERIA]

Read each profile's name and headline carefully. For each one, think about:

• Does this person have deep domain expertise relevant to the company's needs?
• Have they built or scaled companies themselves (operator experience)?
• Do they have a track record of advising, board membership, angel investing, or mentoring?
• Would their industry credibility and network open meaningful doors?
• Does their headline signal willingness to advise (angel investor, advisor, board member, mentor, speaker)?
• Would they fill a genuine skill gap vs. duplicating existing strengths?

Think like a founder assembling a world-class advisory board, not a recruiter filling a role. The best advisors combine relevant experience with a genuine willingness to help.

Mode: vcs

You are a fundraising strategist evaluating investor fit for a startup raise.

Goal: Find the best investor matches for: [CRITERIA]

Read each profile's name and headline carefully. For each one, think about:

• Does this person's fund or firm invest at the right stage and in the right sector?
• Are they a partner or decision-maker, or a junior associate with limited check-writing authority?
• Do their portfolio companies suggest thesis alignment without creating conflicts?
• Does their background suggest they'd add value beyond capital (operational help, intros, domain expertise)?
• Is there anything in their headline that signals active deployment (e.g., "investing in...", "backing...")?

Think like a founder choosing investors they'll work with for 10 years, not a spray-and-pray fundraiser. Fit matters more than fame.

Common instructions (appended to all modes)

Return your top 15 candidates as a JSON array:

[{ "name": "...", "headline": "...", "profileUrl": "...", "reason": "1 sentence why" }]

Be selective. Only include people who genuinely fit. Returning 5 great matches is better than 15 mediocre ones.

End your response with a stats line: "Reviewed [N] profiles. Selected [M] candidates. Top pick: [name] — [reason]."

Between waves (if more than 3 batches), report:

Wave 1/2 complete: [M] candidates from [N] profiles ([B] batches). Starting wave 2...

Step 3: Merge, rank, and report

Collect results from all agents. Deduplicate by profileUrl. Re-rank the combined candidates into a final top 30, assigning tiers based on mode:

Leads tiers

• Tier 1 (ranks 1-12): Strongest fit — contact first
• Tier 2 (ranks 13-22): Strong fit — pursue after Tier 1
• Tier 3 (ranks 23-30): Adjacent — may need qualification

Advisors tiers

• Tier 1 (ranks 1-12): Ideal advisors — reach out immediately
• Tier 2 (ranks 13-22): Strong candidates — worth a serious conversation
• Tier 3 (ranks 23-30): Worth a conversation — explore fit

VCs tiers

• Tier 1 (ranks 1-12): Top targets — reach out first
• Tier 2 (ranks 13-22): Strong fit — pursue actively
• Tier 3 (ranks 23-30): Worth exploring — may need qualification

After ranking, display a completion summary:

Analysis complete

• Profiles reviewed: [total]
• Candidates surfaced by agents: [count]
• Final top 30 selected
• Agents used: [count] ([waves] wave(s))
• Estimated token usage: ~[T]K input, ~[O]K output

Output

Mode-specific labels

Console display

Show a formatted table for each tier:

## [Header] from [person]'s Network
### Criteria: [user's criteria]

### Tier 1 — [tier 1 label]
| # | Name | Headline | LinkedIn | Why |
|---|------|----------|----------|-----|
| 1 | ...  | ...      | url      | ... |

### Tier 2 — [tier 2 label]
...

### Tier 3 — [tier 3 label]
...

Always include the full LinkedIn URL for each person.

File output

Save to ${CLAUDE_SKILL_DIR}/../../output/$0/:

1. [file prefix].json:

[
  {
    "rank": 1,
    "name": "Person Name",
    "headline": "Their LinkedIn headline",
    "profileUrl": "https://www.linkedin.com/in/...",
    "tier": 1,
    "reason": "One-line explanation"
  }
]

2. [file prefix].csv with headers: Rank,Name,Headline,LinkedIn URL,Tier,Reason

Report the saved file paths when done.