linkedin-profile-perfector

name

linkedin-profile-perfector

description

Audit and rewrite a LinkedIn profile into a recruiter-grade report from the user's material (PDF, resume, website, or pasted text). Use when the user asks to optimize, improve, review, or audit their LinkedIn profile, headline, About section, or experience, or wants help with personal branding, LinkedIn for job search, LinkedIn for consulting or clients, or growing a creator profile. Runs a 7-step adaptive intake, scores every section 0-10, applies a recruiter 6-second-scan, produces Current-to-Optimized rewrites, and renders report.md plus report.html. Four goal modes re-weight the pipeline: personal branding, job search, creator growth, and consulting.

linkedin-profile-perfector

Overview

This skill turns your LinkedIn material (PDF export, resume, website, or pasted text) into:

1. A section-by-section audit scored 0–10, with a recruiter 6-second-scan applied across all modes.
2. Copy-paste-ready rewrites for headline, About, every experience role, and skills — formatted as Current → Optimized pairs.
3. A slim, prioritized action checklist tuned to your goal mode.
4. Two output files: runs/<YYYY-MM-DD>-<slug>/report.md and runs/<YYYY-MM-DD>-<slug>/report.html.

Four goal modes — select one in Step A:

Mode	Purpose
1. Personal branding	Build authority; position yourself; About-as-manifesto
2. Job search	Recruiter scan, ATS keywords, searchability, completeness
3. Creator / grow profile	Hook quality, profile-as-landing-page, follower conversion
4. Consulting / win clients	Proof, ROI, services clarity, client-attraction copy

Honest limits: This skill works from the material you provide. It cannot read live LinkedIn metrics (SSI, profile views, post impressions, engagement rate) from a PDF export — those are flagged as "not available — check at linkedin.com/sales/ssi" rather than estimated. It does not auto-post to LinkedIn, generate profile images or banners, or access your LinkedIn account directly. See INSTALL.md for setup instructions and the full honest-limits section.

---

Step A — Run the Adaptive Intake

Load references/intake-questions.md for the full 7-step flow spec, progress-bar render format, and auto-fill algorithm. Then execute the intake as follows.

Before each question, print the progress bar using the exact format defined in references/intake-questions.md:

🟩🟩🟩⏭️🟩⬜⬜   Step 5 of 7
✅ 1. Goal — Job search
✅ 2. Material — PDF + website + resume
✅ 3. Target role — Senior PM  (read from your profile)
⏭️ 4. Industry — B2B SaaS  (auto-filled, skipped)
▶️ 5. Audience …

Legend: 🟩 answered · ▶️ current · ⏭️ auto-filled/skipped (with one-line reason) · ⬜ pending.

For each question, use the AskUserQuestion tool to render it as a dropdown with:

• Options generated from the user's parsed material and context (specific, not generic).
• A "🤖 Recommend for me" option on every question.
• An "Other (type your own)" escape on every question.

The 7 steps:

Step 1 — 🎯 Goal mode (always asked) Question: "What's the #1 thing you want this profile to do?" Options: Personal brand (build authority) · Land a job · Grow as creator · Win consulting clients · 🤖 Help me decide · Other. This answer sets the mode for every subsequent step.

Step 2 — 📥 Material (always asked) Question: "Drop everything you've got — in one place." Multi-select: LinkedIn PDF (provide file path) · Resume/CV (provide file path) · Website URL · Portfolio/project links · Pasted profile text · Personal notes/wiki · "Just describe me — I'll type it here" · 🤖 Recommend what to gather. After the user answers, immediately prompt for the path, URL, or text for each selected type. Ingest all of it before Step 3.

Step 3 — 🧑‍💼 Target role / niche (auto-fill eligible) Frame the question to match the mode:

• Job search → "What title do you want to land?"
• Consulting → "What service or role do you offer?"
• Personal branding / Creator → "What do you want to be known for?"

Generate dropdown options from the material (e.g. current title, titles mentioned in experience). If Step 2 material clearly reveals the answer (high confidence), auto-fill and mark ⏭️ in the progress bar with a one-line reason. Otherwise ask with generated options + 🤖 + Other.

Step 4 — 🏢 Industry / field (auto-fill eligible) Question: "What industry or field is that in?" Skip entirely if the material already shows the industry (high confidence) — mark ⏭️. Otherwise generate options from material signals + 🤖 + Other.

Step 5 — 👥 Target audience Question: "Who most needs to be impressed when they land on your profile?" Generate options tuned to mode + context:

• Personal branding → business owners, industry peers, media/journalists
• Job search → recruiters, hiring managers, department heads
• Creator → niche audience, collaborators, brand sponsors
• Consulting → potential clients, procurement decision-makers, referral partners Always include 🤖 + Other.

Step 6 — ⭐ Top achievements / selling points Question: "What are your top 2–3 achievements or biggest selling points?" Extract top candidates from the material and propose them as pre-selected options. The user picks, edits, and/or adds their own. Always include 🤖 (use what I found in your material) + Other.

Step 7 — 🔑 Keywords to rank for Question: "What keywords should your profile rank for?" Auto-generate a keyword list from goal + role + industry using references/copy-playbooks.md keyword-sourcing rules. Present them as pre-selected options for the user to confirm or adjust. Offer: "Paste a target job description to sharpen this list." Always include 🤖 + Other.

Auto-fill / skip rule (Steps 3–4 in particular): After ingesting Step 2 material, attempt to answer Steps 3, 4, 5, 6, and 7 from the parsed content. If confidence is high, present as a pre-filled confirmation ("I found X in your material — confirm or change?") or mark ⏭️ and proceed. If confidence is low, ask with generated options. Never skip Step 1 or Step 2.

---

Step B — Ingest Material

Read all material the user provided in Step A using the appropriate tool for each type:

• LinkedIn PDF / resume PDF / resume doc → use the Read tool with the exact file path the user provided.
• Website or portfolio URLs → use the WebFetch tool on each URL.
• Pasted text or notes → treat as inline content already in the conversation.

Merge everything into a single working profile model with these fields:

profile_model:
  headline: <string>
  about: <string>
  experience:
    - title, company, dates, description, bullets[]
  skills: []
  education: []
  recommendations: []
  role_signals: []       # inferred job titles, levels
  industry_signals: []   # inferred industries, domains
  audience_signals: []   # inferred audiences from language/context
  achievements: []       # quantified wins, impact statements
  keywords: []           # recurring terms, explicit skill mentions
  pdf_gaps: []           # fields not present in a PDF (SSI, views, etc.)

Note every field that could not be extracted from the provided material — these become audit gaps. Mark any live-only metrics (SSI score, profile views, post impressions, follower count from private data) in pdf_gaps with the label: "not available — check at linkedin.com/sales/ssi".

---

Step C — Classify

Using the working profile model from Step B combined with the intake answers from Step A, determine:

1. Industry — the primary industry or field (e.g. B2B SaaS, Healthcare, Finance, Marketing, Consulting, Creative).
2. Profile type — the user's seniority and function (e.g. IC contributor, manager, executive, founder, freelancer, creator).
3. Target audience — the primary audience the profile needs to impress (confirmed from Step A-5).

Record these three signals. They drive which industry benchmarks are selected in Step D and which copy formulas are prioritized in Step E.

---

Step D — Audit

Load the following three reference files in full before scoring:

• references/scoring-framework.md — 0–10 rubrics per section, weighted overall score, mode weight tables, PDF gap handling rules.
• references/benchmarks.md — industry benchmark ranges; use as context to calibrate the score narrative, never fabricate live metrics.
• references/industry-best-practices.md — All-Star checklist, recruiter visibility rules, recommendations strategy, featured section, content cadence.

Score each profile section 0–10 using the rubrics in references/scoring-framework.md:

Section	Notes
Photo	Score only if image is available; else flag as "not assessable from PDF"
Banner	Score only if image is available; else flag as "not assessable from PDF"
Headline	Always scoreable
About	Always scoreable
Experience	Score each role; flag missing metrics
Skills	Score count, relevance, top-3 pinned
Recommendations	Score count, recency, quality signals
Activity / posts	Flag as "not available from PDF — check linkedin.com" if not provided

Apply the recruiter 6-second scan. The scan is defined canonically in references/industry-best-practices.md §6 (surfaces, the CLICK/SKIP test, and the trigger checklists); references/scoring-framework.md carries only the output format. If the two files differ on the scan, industry-best-practices.md §6 wins. Run it: "In 6 seconds from the headline + first ~300 chars of About + current role, would the target audience click or skip?" Flag skip-triggers (vague headline, no value prop, missing keywords, "seeking opportunities" language) and click-triggers as positives.

Apply mode weights from references/scoring-framework.md to compute the weighted overall score. For example: Job search weights Headline + Experience + Keywords higher; Creator weights About + Activity; Consulting weights About + proof elements.

Calibrate deduction severity with industry context. Before finalizing each section score, check references/benchmarks.md for the profile's industry: the same gap can be a red flag in one field and normal in another (e.g. "2 recommendations is weak in consulting but average in finance"; "generic recommendations are not a red flag in finance"). Soften or sharpen the deduction accordingly, and name the industry context in the scorecard's Top Fix column so the user understands why a gap was or wasn't penalized.

Flag every gap found: missing sections, weak copy, absent metrics, keyword gaps, incomplete education or skills, no recommendations.

Mark live-only metrics as: not available — check at linkedin.com/sales/ssi.

---

Step E — Rewrite

Load references/copy-playbooks.md in full before writing any copy.

Anti-fabrication guardrail (highest priority — see copy-playbooks.md §1.0). Never introduce a metric, employer, job title, date, client name, dollar figure, or achievement that is not present in the user's material. Optimize how true facts are expressed; do not invent new ones. Where a stronger line clearly needs a number the source lacks, insert a bracketed placeholder ([add metric: …], [add client name], [add timeframe]) for the user to fill. This overrides every other copy rule.

Produce Current → Optimized pairs for every rewriteable section. Use the formulas, examples, and per-mode variant tables in references/copy-playbooks.md. Tune emphasis to the selected mode:

Headline:

• Format: [Role] | [Key Expertise] | [Value Prop] (max 220 chars).
• Job search → keyword-dense, ATS-optimized, target title prominent.
• Creator → value prop for followers, what they get by following you.
• Consulting → outcome-led, client result prominent.
• Personal branding → point-of-view, what you stand for.

About:

• Limit: ~1,500–2,000 chars (2,600 max).
• Structure: hook (first ~300 chars must work as "see more" preview) → who you are → achievements → what you offer or seek → CTA.
• Tune narrative voice and CTA to the selected mode.

Experience (every role):

• Format: title · company · dates · 2–3 line summary + 4–6 short bulleted achievements.
• Each bullet: active verb + result/metric + keyword where natural.
• Job search → metric-driven, keyword-rich, ATS terms visible.
• Consulting → outcome and ROI visible in bullet structure.
• Show Current: and Optimized: blocks for each role.

Skills:

• Up to 50 skills, top 3 pinned.
• Align to keywords from Step A-7 and target job/audience language.
• Remove vague or outdated terms; add keyword-relevant gaps.

For every rewrite, apply the embedded copy rules from references/copy-playbooks.md: short & concise · keyword-rich · bullet every experience role · benefits over features · active voice · metrics where possible · cut weak intensifiers ("really", "very", "just") · strip AI-filler ("I hope this finds you well", "leverage", "synergy", "passionate about").

Project the optimized score (before → after). After producing the rewrites, re-score the Optimized copy against the same rubrics and mode weights from references/scoring-framework.md used for the current audit in Step D. This yields a projected score per section and a projected overall score. Rules:

• The projected score reflects only the rewritten copy — it must be defensible from what the Optimized version actually says. Do not inflate it, and never count a metric the rewrite only left as a [add metric: …] placeholder (see the anti-fabrication guardrail above). A section whose lift depends on the user filling placeholders should note that.
• Always label projected numbers "(projected)" so they read as an achievable target on the same rubric, not a promise about LinkedIn's algorithm.
• Carry both numbers forward to Step G as now → projected for the overall score and each scorecard row.

---

Step F — Action Checklist

Produce a slim, mode-specific, prioritized action checklist — no more than 10–12 items. Prioritize by impact on the selected goal mode. Tie each item to a specific audit gap or rewrite from Steps D and E.

Mode framing:

• Job search → recruiter visibility, Open-to-Work settings, target-role keyword gaps, recommendation count, completeness for ATS.
• Personal branding → content pillars, signature topics, featured section, engagement habits.
• Creator / grow → posting schedule, hook formula, profile-as-landing-page, CTA in About.
• Consulting → services section, proof/case-study copy, lead magnet, social proof, outreach.

---

Step G — Render

Fill the templates with all content from Steps D, E, and F. Before substituting, compute these tokens — they are easy to miss because they control color and structure, not just text:

Score → color class (HTML only). The template no longer hardcodes a color; you must set the class token from the actual score so the badge color matches the number:

Token	Set from	Value
{{SCORE_CLASS_BEFORE}}	current overall score	score-green if ≥75 · score-amber if 50–74 · score-red if <50
{{SCORE_CLASS_AFTER}}	projected overall score	same thresholds as above
{{SCAN_CLASS}}	recruiter scan verdict	scan-click if CLICK · scan-skip if SKIP

Before → after scores (from Step E projection):

• {{OVERALL_SCORE_BEFORE}} and {{OVERALL_SCORE_AFTER}} — the current and projected overall (out of 100).
• Each {{SCORECARD_ROWS}} row must include both the Now and Projected section scores (0–10), e.g. Headline | 4 | 9 | ….

Keyword gap table {{KEYWORD_GAP_TABLE}} — one row per target keyword with ✅/⬜ for whether it currently appears in each surface, using the placement priority in references/copy-playbooks.md: keyword | Headline | About | Experience | Skills.

Profile infrastructure {{PROFILE_INFRA_CHECKLIST}} — a checklist covering custom URL, banner, Featured section, Open-to-Work / Providing-services settings, and recommendations strategy. Items not assessable from the material are marked accordingly rather than scored.

Then:

1. Read assets/report-template.md — fill every {{PLACEHOLDER}} token with the corresponding audit scores, rewrites, checklist items, and metadata. Write the result to: runs/<YYYY-MM-DD>-<slug>/report.md where <YYYY-MM-DD> is today's date and <slug> is a short kebab-case name derived from the person's name or target role (e.g. alex-chen, senior-pm, marketing-consultant).

2. Read assets/report-template.html — fill every {{PLACEHOLDER}} token identically. Write the result to: runs/<YYYY-MM-DD>-<slug>/report.html

3. Verify both output files have zero unfilled {{...}} tokens before reporting done. If any remain, fill them or replace with a clearly labeled "not available" note.

4. Tell the user the exact paths to both files:

   • runs/<YYYY-MM-DD>-<slug>/report.md
   • runs/<YYYY-MM-DD>-<slug>/report.html

5. Ask: "Want me to open the HTML report in your browser?" If yes, run open runs/<YYYY-MM-DD>-<slug>/report.html.