| name | writer-verification |
| description | Use when existing writing needs reviewing, proofreading, or improvement. Triggers on: "글 검토해줘", "맞춤법 확인", "proofread", "이 문장 자연스러워?", "글 다듬어줘", review my writing", "출판 전에 확인해줘", "표현이 어색한 것 같아", "문법 맞아?", "글 퀄리티 봐줘". |
| scenarios | ["이 이메일 자연스러워?","블로그 포스트 올리기 전에 검토해줘","Proofread this PR description","문서 표현이 어색한 것 같아, 봐줘","이 문장들 맞춤법이랑 스타일 체크해줘","Does this read naturally to a native speaker?"] |
| compatibility | {"optional":["think-tool"],"remote_mcp_note":"think-tool이 있으면 어떤 발견이 독자 경험에 가장 큰 영향을 미치는지 종합할 때 활용됩니다. Claude 설정 → MCP Servers에서 remote SSE 엔드포인트를 추가하세요."} |
Writer Verification
Core rule: Every finding must include original → fix + reason. Pointing out without fixing is half a review.
MCP usage by mode:
- Inline mode only (< 300 chars): If sequential-thinking is available, use it to structure the four passes as sequential steps.
- Both modes: If think-tool is available, use it before each pass to reason through the text and audience before acting.
- Parallel mode, before Summary: If think-tool is available, invoke it before writing the Summary section to identify which findings most affect the reader's experience.
Execution Mode
Count the characters in the submitted text first.
< 300 chars — inline: Read references/passes.md and run all four passes sequentially. If agents/*.md are updated, update references/passes.md in sync — they are the inline-mode mirror.
≥ 300 chars — parallel subagents: Dispatch all four in the same turn. Wait for all to return, then aggregate.
| Pass | Agent | Inputs |
|---|
| 1. Spelling & Grammar | agents/grammarian.md | text, language |
| 2. Writing Patterns | agents/editor.md | text |
| 3. Expression & Style | agents/copywriter.md | text, purpose, audience |
| 4. Reader Perspective | agents/reader.md | text, audience |
If a Korean spelling issue is uncertain, load references/korean-spelling.md.
Aggregation rules (parallel mode):
- Deduplicate by location: if two passes flag the same span, keep the higher-severity finding and note both sources
- Conflicting severity: elevate to the higher level (🔴 beats 🟡 beats 🟢)
- Conflicting fixes: surface both options with attribution (e.g.,
[grammarian] fix A / [copywriter] fix B)
- If mcp-reasoner is available, invoke it during aggregation to resolve ambiguous conflicts before producing the final report (Ambiguous = two passes flag the same span for incompatible reasons, e.g., grammarian says rewrite for subject agreement while copywriter says cut the sentence entirely for vagueness — invoke mcp-reasoner to choose)
Output Format
See references/output-format.md for the full template and examples.
Priority: 🔴 Must fix (meaning errors, logic gaps) · 🟡 Recommended (patterns, phrasing) · 🟢 Optional (style preference)
