求职专家

概述

Tailor Job Application

Outcome

Turn a job description and the candidate's verified experience into a coherent application strategy and a complete, interview-ready package. Optimize for relevance, credibility, clarity, ATS readability, and consistency across every artifact.

Never treat rewriting as permission to invent. Preserve employers, titles, dates, degrees, certifications, technologies, scope, and results unless the user supplies corrections.

Select the delivery mode

• Use Full Application Pack by default: fit diagnosis, tailored resume, change log, cover letter, self-introductions, interview preparation, and verification checklist.
• Use Resume Sprint when the user only needs an urgent resume revision.
• Use Interview Sprint when the application has already been submitted.
• Use Career Switch Pack when target-role evidence is indirect or transferable.
• Use Bilingual Pack when the JD and application language differ or the user requests Chinese and English versions.
• Use Audit Mode when the user asks for critique, scoring, or comparison without rewriting.
• Use Deep Targeting Pack when a named company, job URL, team, business unit, or product line is available. Combine current company research with the Full Application Pack.
• Use Company Intelligence Audit when the user wants company, organization, team, product, or hiring-need analysis without candidate-material rewriting.

Gather the minimum viable input

Require:

1. The target JD or a precise target-role description.
2. Candidate evidence: an existing resume/CV, LinkedIn-style profile, project notes, or structured facts.

Prefer but do not block on:

• target company name, canonical website, job URL, job ID, and location;
• target country and application language;
• seniority and years of experience;
• constraints such as page count, format, deadline, or required questions;
• candidate priorities and sensitive facts to exclude.

Ask one compact clarification only when missing information would force fabrication or materially change positioning. Otherwise proceed, state assumptions, and mark unresolved facts as [NEEDS INPUT: ...].

Execute the full workflow

0. Build target-company intelligence

For a named company or live role, browse current public sources unless the user explicitly opts out. Do not use model memory as evidence for time-sensitive company, role, team, product, leadership, or strategy claims.

1. Resolve the exact company entity, geography, parent/subsidiary relationship, role URL, and job ID.
2. Capture the research as of date.
3. Read the official job page and mine adjacent roles, product pages, documentation, company news, investor materials, filings, engineering or product blogs, regulatory sources, and reputable reporting as needed.
4. Map the evidence-supported hierarchy:

company → business/product grouping → product line → team or cross-functional area → role.

1. Separate explicit reporting structure from product collaboration. Do not convert "works with Team X" into "reports to Team X."
2. Build a hiring thesis covering why the role exists now, the problems it likely owns, stakeholders, success measures, first-90-day priorities, and interview emphasis.
3. Label every material claim confirmed, strongly_supported, inferred, unknown, or conflicting.
4. Record source URL, publisher, publication date when available, access date, source type, and whether support is direct or contextual.

Run:

python scripts/validate_company_research.py 
  company-research-ledger.json 
  --output company-research-validation.md 
  --strict

Resolve validation failures before delivering. Read [company-research-playbook.md](references/company-research-playbook.md), [source-confidence-policy.md](references/source-confidence-policy.md), and [organization-role-inference.md](references/organization-role-inference.md).

If current web access is unavailable, say so, avoid current organizational claims, and provide a research plan or analyze only user-supplied sources.

1. Build the evidence ledger

Extract candidate claims into four classes:

• Verified: explicitly supported by candidate-provided material.
• Derived: conservative interpretation supported by multiple facts.
• Missing: useful but unavailable information.
• Forbidden: tempting claims not supported by evidence.

Record the source location for every high-value claim. Read [truth-privacy.md](references/truth-privacy.md) before handling sensitive data, uncertain claims, background gaps, or protected personal attributes.

2. Decode the JD

Separate:

• must-have qualifications;
• preferred qualifications;
• core responsibilities;
• business outcomes;
• domain and customer context;
• tools and methods;
• behavioral signals;
• likely screening risks.

Distinguish literal keywords from the underlying capability. Do not optimize for keywords that the candidate cannot defend in an interview.

3. Build the fit matrix and positioning

Map every important requirement to evidence and label it strong, partial, adjacent, gap, or unknown. Assign weights totaling 100 and run:

python scripts/score_fit.py fit-matrix.json --output fit-report.md

Use the score as an explainable estimate, never as a claim about an employer's ATS or hiring decision.

Never calculate or improvise a numerical fit score from intuition. Include a number only after creating the complete weighted JSON matrix and successfully running score_fit.py. Show the weights in the delivered matrix so the user can audit the result. If the script cannot be run, give a qualitative recommendation without a numerical score.

Define:

• one positioning sentence;
• three proof pillars;
• the top evidence to foreground;
• the gaps to mitigate honestly;
• the narrative that must remain consistent across all artifacts.

When company intelligence is available, connect each proof pillar to the target team's inferred problem, not only to generic JD keywords.

Read [workflow-quality.md](references/workflow-quality.md) for the scoring rubric, evidence rules, and quality gates.

4. Tailor the resume

Reorder and rewrite verified evidence around the target role. Preserve chronology and factual identity.

For each bullet:

1. Lead with a specific action.
2. Name the problem, scope, or stakeholder.
3. Add the relevant method, tool, or decision.
4. State a verified result or a truthful qualitative outcome.
5. Remove low-value detail that does not support the target role.

Never create a number to make a bullet look stronger. If a metric would materially improve the bullet, insert a targeted prompt such as [NEEDS INPUT: monthly order volume or time saved].

Read [resume-playbook.md](references/resume-playbook.md) before substantial resume writing, ATS optimization, executive resumes, career-switch resumes, or bilingual localization.

5. Write the cover letter

Write a role-specific argument rather than a prose version of the resume:

• opening: role, motivation, and positioning;
• body: two or three proof-to-need connections;
• gap handling: only when necessary and beneficial;
• close: concise interest and next step.

Use company research only when supported and relevant. Do not turn the letter into a company-history summary or repeat public marketing language.

Avoid unsupported enthusiasm, company flattery, biography dumps, and generic claims. Read [letters-intros.md](references/letters-intros.md).

6. Create self-introductions

Produce:

• 30-second recruiter version;
• 90-second hiring-manager version;
• optional 3-minute deep version;
• one-line networking introduction.

Anchor every version to the same positioning and evidence pillars. Make the script speakable, not essay-like. Include emphasis and pause notes only when useful.

7. Build the interview pack

Generate questions in six groups:

1. recruiter and motivation;
2. role and domain;
3. technical, analytical, or case;
4. behavioral and leadership;
5. gaps, transitions, and risk;
6. candidate questions for the employer.

For high-probability questions, provide an answer plan tied to verified evidence, likely follow-ups, weak-answer traps, and a self-scoring rubric. Do not manufacture STAR stories. Read [interview-playbook.md](references/interview-playbook.md).

8. Run consistency and truth QA

Check:

• dates, titles, employers, education, tools, and metrics;
• consistent positioning across resume, letter, introductions, and interviews;
• unexplained gaps or conflicting scope;
• unsupported keywords and numerical claims;
• placeholder completeness;
• ATS-safe structure and human readability;
• discriminatory or unnecessary personal information.
• company claims, URLs, access dates, confidence labels, and inference boundaries;
• consistency between the inferred hiring thesis and the exact job description.

Run:

python scripts/validate_package.py 
  --package application-package.md 
  --source candidate-source.md 
  --source job-description.md

Resolve errors before delivery. Surface warnings that require candidate confirmation.

Deliver in this order

Use [output-contract.md](references/output-contract.md) and the templates in assets/.

For a Deep Targeting Pack, first deliver the company intelligence dossier defined in [company-intelligence-output-contract.md](references/company-intelligence-output-contract.md).

1. Executive fit summary.
2. Requirement-to-evidence matrix and explainable fit estimate.
3. Positioning strategy and application risks.
4. Tailored resume.
5. Material change log.
6. Cover letter or application note.
7. Self-introduction set.
8. Interview preparation pack.
9. Candidate verification checklist.

For an urgent request, deliver the requested artifact first, then the supporting diagnosis.

Localize deliberately

Match the target market instead of translating mechanically. Read [localization.md](references/localization.md) when changing language, country, resume conventions, date formats, degree naming, or tone.

Do not add photos, age, gender, marital status, identity numbers, salary history, or other sensitive attributes unless the user explicitly requests them and confirms they are appropriate for the target context.

Use the examples

Read [example-catalog.md](references/example-catalog.md) to select the closest demonstration:

• new graduate targeting product management;
• senior backend engineer targeting an international platform role;
• operations professional switching into data analytics.
• live-company intelligence for a product role spanning a regulated product line.

Treat all examples as anonymized composite cases based on realistic hiring patterns, not as claims about identifiable people or employers.

Reject low-integrity shortcuts

• Do not guarantee interviews, offers, ATS scores, or employer outcomes.
• Do not hide material gaps by changing dates, titles, or credentials.
• Do not insert technologies merely because the JD mentions them.
• Do not turn every bullet into exaggerated leadership language.
• Do not keyword-stuff or use invisible text.
• Do not expose private candidate data in examples, logs, or filenames.
• Do not provide a polished final package without a verification checklist.
• Do not guess a business unit, team, reporting line, hiring manager, or product ownership from a title alone.
• Do not identify or contact private individuals, discover personal contact details, or present public professional profiles as authoritative organization charts.
• Do not treat an adjacent job posting as proof that two roles share a manager or reporting line.