linkedin-profile-optimizer
Audit or rewrite LinkedIn profiles and role keywords from verified evidence; not CVs, letters, feed posts, or asset files.
SKILL.mdcodex/skills/linkedin-profile-optimizer/SKILL.md
GitHub<!-- GENERATED BY generate_skill_wrappers.py; DO NOT EDIT MANUALLY. -->
Generated Codex skill — do not edit manually
This complete skill and its bundled resources are generated from agent-plugins/skills/linkedin-profile-optimizer/SKILL.md. Manual changes in this directory will be overwritten the next time the generator runs. Edit the canonical skill instead.
LinkedIn Profile Optimizer
Make the profile a coherent path from search to relevant expertise, evidence, and a useful next action. This is a reader journey, not a claim about LinkedIn's ranking algorithm. Preserve the person's actual work and voice instead of copying a CV.
Establish scope and context
- Identify the requested operation: audit, rewrite, keyword map, or a named section. An audit alone does not authorize a rewrite. Keep a headline request narrow.
- Establish the primary goal (employment, clients, or professional networking), audience, role family, seniority, language, and geography from supplied context. Ask only for missing decisions that materially affect the result, in one bundle.
- Read current project instructions, available relevant memory, candidate-source configuration, and personal voice rules. Use memory to locate sources and prior decisions; verify candidate claims against current evidence. Do not edit memory unless explicitly requested. If context is unavailable, name the limitation.
- Accept pasted profile text, export, screenshots, or accessible supplied URLs.
A handle alone proves no profile content. Attempt permitted access when useful;
if unavailable, request text or an export and continue only with shown sections.
Mark unseen sections
not reviewed, never failed or absent. - For search alignment, use explicitly supplied target postings or a user-authorized selection. A single posting supports a role-specific hypothesis; several distinct postings can support a broader pattern. Record sample size and selection bias.
Read evidence and project integration before using career evidence or job-intelligence data. Read platform claims and provenance before stating current limits, taxonomy, Recruiter behavior, or algorithm advice.
Build positioning and the keyword map
- Form one evidence-backed positioning sentence: target work, recurring problem, candidate contribution, and strongest available proof. Use the recurring thread across roles without inventing a mission, preference, or causal impact.
- Separate required capabilities, preferred capabilities, domain, role/title, location/language constraints, and generic low-signal wording. Keep each posting's exact excerpt and source; do not turn repeated text from duplicate ads into demand.
- Build rows with
term,source_id,jd_quote,importance,basis,candidate_quote,evidence_id,match,placement, andtaxonomy_status. Usestated|structural|inferredfor basis andstrong|partial|missing|unknown|not_applicablefor match. Distinguish a documented absence from unavailable evidence. Do not treat a fit score as evidence of skill. - Prefer keywords that matter to the selected work and have candidate evidence. A missing desirable skill belongs in a gap or learning note, not public copy. Keep synonyms for human clarity without claiming every synonym is a selectable tag.
- For backend candidates, assess OpenTelemetry/OTel, eBPF, distributed-system resilience, idempotency, and gRPC only when relevant and evidenced. They describe different capabilities; none substitutes for another. Monitoring does not prove OpenTelemetry, Linux does not prove eBPF, and REST does not prove gRPC. Retain core technologies such as Go or PHP when central to the role. Do not promise priority in Recruiter filters for niche terms.
- Verify exact selectable Skills labels in current official material or an observed
skill picker. Otherwise set
taxonomy_status: not_verified; suggest checking the picker and retain a truthful phrase in Experience where useful. Do not save changes simply to test taxonomy. Record the date and surface of any actual observation.
Audit and write only the requested sections
Read section playbook for audit criteria and section-specific drafting. Review nine surfaces when shown: photo, banner, headline, About, Featured, Experience, Skills, public URL, and recommendations.
- Report
supported,needs revision, ornot reviewedwith concrete evidence. Separate missing facts, weak presentation, and inaccessible UI. Rank fixes by audience relevance, clarity, proof, and effort; do not invent numerical uplift. - Make the headline searchable and specific. A useful starting shape is
Role | Specialism | Supported contribution; omit an unsupported result rather than filling the space. Count the final characters, including spaces, against the currently verified limit or a clearly labeled provisional budget. - Write About in first person with a useful opening, concrete work, selected proof, working approach, and goal-appropriate invitation. Treat seven steps and a roughly 270-character hook as optional editing aids. Inspect actual mobile/desktop truncation when available; never promise a fixed universal “see more” boundary.
- In Experience, connect problem, personal action or decision, scope, and supported outcome. Preserve job titles, dates, team versus individual ownership, and uncertainty. A qualitative result is valid; every bullet need not have a metric.
- Prioritize a small set of strongest evidenced skills overall and relevant skills per role. Suggest a top three as editorial priorities only where useful; verify current UI support for ordering, pinning, and associations before prescribing it.
- Suggest public repository, release, diagram, or case-study links at the relevant role's supported media/link surface. Skills may associate with roles; do not assume a skill card supports arbitrary external links. Verify authorship and contribution, not merely that a repository exists or uses the technology.
- For Featured, propose up to three complementary items when the evidence warrants
them: expertise overview, substantial work sample, and goal-specific next step.
A smaller strong set is preferable to filler. Use
linkedin-featured-proofif available when the user requests asset files; otherwise follow the requested document workflow without requiring that sibling skill. - Draft recommendation requests only if requested, around a shared project and facts the recipient can attest to. Never fabricate a testimonial or send it.
Deliver and verify
Return the smallest complete package for the selected operation:
- Audit: observed coverage, evidence-linked findings, prioritized changes, uncertainties, and questions that block specific decisions.
- Rewrite: paste-ready requested sections, final headline count, ordered Skills with evidence and taxonomy status, proposed proof links and CTA where in scope.
- Keyword map: sampled sources, anchored terms and candidate matches, placement plan, unsupported gaps, and clearly separated editorial inference.
Keep the claim ledger and keyword map available as review notes; expose full matrices only when requested or needed to explain uncertainty. Keep source IDs, internal review labels, and private excerpts out of public prose. Create files only when requested. Check every final claim, link, translation, and voice against the same evidence; count final text again after edits. Explain material discrepancies instead of silently choosing a source. Do not hide audit findings when requested.
Deliver drafts without changing LinkedIn, visibility, public URL, job-search settings, messages, or other external state unless explicitly authorized for that operation. Existing authorization need not be requested again. Record what was actually changed only after visible confirmation; preparation is not publication.
If measurement is requested, establish a dated baseline and track relevant search appearances, qualified inquiries, or portfolio visits where available. Label before/after observations as observational; do not attribute causality to a rewrite.