| name | ja-so-seo-do-u-mi |
| version | 0.2.0 |
| description | Use when the user wants to revise, diagnose, rewrite, shorten, lengthen, or polish a Korean job application essay or 자기소개서 using a pasted 채용공고/JD, 자기소개서 문항, 글자수 제한, and rough draft. Default to preserving the user's intended appeal points unless they explicitly ask for a full rewrite. Trigger on requests such as "자소서 수정", "자기소개서 고쳐줘", "채용공고에 맞게 자소서", "JD 기반 자소서", "문장만 다듬어줘", "글자수 맞춰줘", "AI 티 안 나게 자소서", "기업 제출용으로 다듬어줘", and "ja-so-seo-do-u-mi". |
자소서 도우미
Revise Korean job application essays by reading the JD, prompt, length rule, and draft together. Default to preservation-first editing: improve structure, evidence, naturalness, and submission safety without silently dropping what the user wanted to appeal.
Input Contract
Accept a simple paste when these four parts are present or inferable:
채용공고/JD:
자소서 문항:
글자수:
초안:
Optional signals: company, role, desired tone, requested mode, must-preserve facts or appeal points, blind-hiring constraints, and length-counting basis.
Ask one short question only when a required part is missing, multiple prompts cannot be mapped, or a must-preserve point conflicts with prompt fit, length, fabrication safety, interview safety, or blind-hiring rules.
Core Rules
- Never invent experiences, numbers, achievements, tools, company facts, awards, periods, or responsibilities.
- Preserve user-provided facts and intended appeal points unless the user explicitly changes them or a hard submission risk requires compression or omission.
- Default to
preserve_revision; use rewrite only for explicit requests like "아예 새로", "구조를 갈아엎어", or "제출본으로 재작성".
- Answer the application prompt before polishing prose.
- Use the JD as context, not a keyword script; responsibilities and requirements outrank preferred qualifications and generic values.
- Replace vague traits with concrete behavior, role, result, and JD relevance without deleting the user's intended strength.
- Remove formulaic application prose only after checking factual meaning and Appeal Lock.
- Stay within the length limit and report residual risks, including any locked appeal point that was heavily compressed or omitted.
References
Load only what the task needs:
references/intake-schema.md: parse input, requested mode, appeal points, and clarifying-question rules.
references/jd-parser-rules.md: extract company, role, responsibilities, requirements, skills, and values.
references/question-taxonomy.md: classify prompt type and required answer elements.
references/field-writing-rules.md: recruiter-style writing heuristics and concrete evidence principles.
references/diagnosis-taxonomy.md: diagnose weak drafts, including Intent Loss Risk.
references/application-naturalness-rules.md: reduce formulaic Korean application prose without erasing intent.
references/rewrite-playbook.md: plan preserve/rewrite modes, length strategy, and final prose.
references/audit-checklist.md: audit factual fidelity, user-intent fidelity, length, prompt fit, JD fit, and submission safety.
scripts/count_korean_length.py: deterministic length and byte counting.
Workflow
- Intake: split or infer JD, prompt, length rule, draft, requested mode, constraints, and confidence. Use
intake-schema.md.
- Appeal Lock: before deep JD parsing or rewriting, lock the user's main claim, early/repeated/detailed experiences, intended strengths, values, motivation, and explicit "꼭 살려줘" points.
- JD Parse: extract job signals with this weight:
responsibilities > requirements > skills/domain > preferred > values.
- Prompt Classification: classify the question and derive mandatory answer elements.
- Diagnosis: detect prompt mismatch, weak JD connection, vague traits, weak action/result, unclear contribution, JD overfit, formulaic prose, fabrication risk, and intent loss risk.
- Rewrite Plan: choose a structure that starts from the prompt answer and keeps locked appeal points unless prompt fit or submission safety requires otherwise.
- Rewrite: produce final submission prose. In
preserve_revision, use the smallest structural change that solves the issue. In rewrite, restructure more freely but preserve or report locked appeal points.
- Naturalness Pass: remove cliches, generic praise, mechanical STAR, repetitive connectors, translationese, exaggerated emotion, and empty ambition while preserving factual meaning and user intent.
- Audit: check factual fidelity, user-intent fidelity, prompt fit, JD fit, length, fabrication risk, and blind-hiring/submission risks. Revise once for hard failures.
Final Response
Default:
[최종 제출본]
글자수: N/M자 (기준: 공백 포함|공백 제외|byte|불명확)
살린 핵심:
- ...
수정 요약:
- ...
축약/생략한 부분:
- ...
확인 필요:
- ...
Include 살린 핵심 by default with 2-4 bullets. Include 축약/생략한 부분 only when content was actually heavily compressed or omitted. Omit 확인 필요 when there is no important residual risk.