Job-tailored resume pipeline

# Job-tailored resume pipeline

AI Summary

Purpose:

• Define the shared Claude Code and Codex workflow for turning a supplied job

posting into a verified, company-specific resume package.

Key points:

• skills/tailor-resume-to-job/SKILL.md is the executable procedure.
• AGENTS.md makes the procedure mandatory for all repository agents.
• The pipeline captures the posting, maps requirements to evidence, selects

relevant modules, uses the canonical template, and runs three persona reviews.

• Publication requires a deterministic document-quality score of at least 95,

CEO/CTO/tech-lead scores of at least 95 each, and zero factual or artifact hard-gate violations.

• The score is not a hiring prediction and cannot be raised through unsupported

claims.

Relevant when:

• A user supplies a job posting or asks for a company-specific resume.
• Auditing why two agents produced different resume packages.
• Diagnosing a failed 95-point release gate.

Do not read full document unless:

• The pipeline architecture, artifacts, or score semantics need explanation.

Linked documents:

• AGENTS.md
• skills/tailor-resume-to-job/SKILL.md
• skills/tailor-resume-to-job/references/quality-rubric.md
• ai/templates/resume-ats-template.html
• ai/assets/resume/README.md
• ai/sources/career/2026-07-22-cross-agent-job-tailored-resume-pipeline-request.md

Open Questions

• The first live application should be used to calibrate whether the three

reviewer prompts are too correlated despite independent inputs.

Details

Control flow

job posting
  -> immutable posting capture
  -> requirement map
  -> verified evidence selection
  -> canonical resume and portfolio assembly
  -> CEO / CTO / tech-lead independent reviews
  -> artifact validation
  -> deterministic score calculation
  -> PASS or exact failed checks

Source discipline

Agents start with the person and project indexes, then open only pages selected by the posting. Resume claims are treated as projections of the evidence map. Wording may change; the metric, role boundary, technology boundary, and source may not drift.

Release contract

Every application package contains the posting, requirement map, Markdown draft, canonical HTML, PDF, review manifest, and quality-gate report. Future packages use the agent-neutral Drive path career/맞춤이력서/<company>-<role>/.

The deterministic score has five categories:

1. evidence integrity: 25
2. posting alignment: 25
3. narrative and role clarity: 20
4. ATS and human readability: 15
5. artifact integrity: 15

Persona review is an additional gate rather than an average that can hide a weak reviewer. Each persona must score at least 95 and have no blocking issue.

Failure behavior

The pipeline stops as DONE_WITH_CONCERNS or NEEDS_CONTEXT when it cannot reach 95 without inventing evidence. It reports the missing source, unsupported posting requirement, or artifact defect. It never weakens evidence checks to obtain a passing number.