24009643/codex-skills

Resume Evidence

Treat a resume as a compact evidence document. The target role changes what is relevant, never what is true.

Scope

Use for resume audits, evidence completion, role or JD tailoring, career changes, layout-only repair, and verified DOCX/PDF delivery.

Do not use for application-form submission, recruiter outreach, portfolio planning, general career plans, or interview-question generation. Route those to separate workflows.

Workflow

1. Lock sources and target. Use the current resume as baseline. Treat explicit answers as user-confirmed, other records as candidate evidence, and the JD only as relevance guidance. Ask the target role only when its absence changes the result; never infer it from this Skill or a template.
2. Audit claims. Track source, personal responsibility, scope or audience, action, evidence and maturity, and uncertainty. Preserve factual identifiers and measurement context. Do not force recurring duties, regulated work, service delivery, craft work, or research into a project narrative.
3. Fill useful gaps. Read CARL evidence and questioning. Ask one direct question at a time, at most three per experience item. Accept “不知道/跳过” and write conservatively.
4. Adapt without stereotyping. Read role adaptation and layout verification. Derive emphasis and section order from evidence, target role, career stage, channel, and JD. Never import a typical skill, credential, responsibility, metric, or outcome. Never present a custom keyword formula as ATS pass probability.
5. Show changes. Prefer location / original / revision / reason / evidence state over a silent rewrite. Reorder or compress only verified material.
6. Format and verify. Use the neutral templates in assets/resume-templates/; show both single-column options when choice matters and default to clean. Render every final DOCX/PDF, inspect every page, compare extracted reading order, and report remaining evidence gaps.

Claim Boundaries

• Separate personal action from team output; use causal or ownership verbs only when supported.
• Accept quantitative and non-quantitative evidence. Never invent numbers because a role is commonly measured numerically.
• Distinguish responsibility, delivery, review or evaluation, observed use, attributed outcome, plan, and unknown.
• Never upgrade a draft, course assignment, simulation, prototype, internal exercise, controlled test, or completed task into external publication, formal deployment, adoption, revenue, learning gain, compliance clearance, or business impact.
• Do not inject occupation-coded language unless the baseline and target evidence support it.
• In layout-only work, content and order are immutable unless separately authorized.
• Do not infer licenses, security clearance, language level, awards, employment status, or protected personal characteristics.
• Do not move, upload, save to a portal, or submit a resume without explicit authorization.