| name | editorial-pipeline-en |
| description | A six-stage editorial quality control pipeline (3 reviews + 3 proofreadings) for blog posts, newsletters, technical articles, and long-form content. Each reviewer role MUST be dispatched as an independent subagent to ensure uncontaminated perspectives. Use when you need rigorous, multi-pass quality assurance on written articles before publication. |
Editorial Pipeline — Article Quality Control
⚠️ Core Principle: Each reviewer role = an independent subagent
Every reviewer must be invoked as a separate subagent — never have the same agent play multiple roles.
This guarantees each review pass has a truly independent perspective, uncontaminated by previous rounds.
When to Use
- Before publishing blog posts, technical tutorials, or newsletters
- Before submitting long-form essays, RFCs, product announcements, or brand copy
- Any high-visibility content where argument rigor and precise language matter
Process Overview
Draft produced
│
▼
[THREE-REVIEW STAGE] — Content-level assessment
├── Review 1 subagent: Structure & completeness
├── Review 2 subagent: Logic, accuracy & depth
└── Review 3 subagent: Final verdict & publish readiness
│
▼ (each round: feedback → revise → next round)
│
[THREE-PROOFREAD STAGE] — Detail-level correction
├── Proof 1 subagent: Line-by-line scan for errors
├── Proof 2 subagent: Verify corrections, catch new issues
└── Proof 3 subagent: Author cold-read, final sign-off
│
▼
Published
Three-Review Specs
Review 1 subagent — content-first-reviewer
Role: Gate basic quality. Prevent obvious holes from reaching later stages.
Checklist:
Output format:
Review 1 Result (content-first-reviewer)
Status: [Pass / Revise]
Issues:
- [Critical] ...
- [Suggestion] ...
Revision instructions: ...
Review 2 subagent — content-deep-reviewer
Prerequisite: Only start after Review 1 passes.
Role: Deep-dive into logical consistency, factual accuracy, and potential risks.
Checklist:
Output format:
Review 2 Result (content-deep-reviewer)
Status: [Pass / Revise / Needs structural discussion]
Major issues: ...
Minor issues: ...
Strengths (worth keeping): ...
Review 3 subagent — content-chief-reviewer
Prerequisite: Only start after Review 2 passes.
Role: Final verdict on publishability. Do not re-litigate details already covered; focus on big-picture judgment.
Checklist:
Output format:
Review 3 Verdict (content-chief-reviewer)
Decision: [Approved / Revise / Hold]
Reasoning: ...
Conditions (if any): ...
Three-Proofread Specs
Proofreading applies only to content that has passed all three reviews. Focus on surface errors, not structural issues.
Proof 1 subagent — proofreader-first
Role: Scan paragraph by paragraph against the final revised draft.
Checklist:
Output format:
Proof 1 Report (proofreader-first)
Error list (with paragraph or line reference):
Para 3: "occured" → "occurred"
...
Proof 2 subagent — proofreader-second
Prerequisite: Proof 1 corrections have been applied.
Role: Independently verify that Proof 1 fixes were applied correctly and catch anything missed.
Checklist:
Output format:
Proof 2 Report (proofreader-second)
Proof 1 fix confirmation: [All correct / X items incorrect]
Incorrect items: ...
Newly found issues: ...
Proof 3 subagent — proofreader-final
Prerequisite: All Proof 2 items resolved.
Role: Final read-through as the author/publisher, simulating the reader experience.
Checklist:
Output format:
Proof 3 Result (proofreader-final)
Status: [Final — cleared for publication / Pending items]
Pending (if any): ...
Sign-off: proofreader-final @ {date}
How to Dispatch Subagents
Every review and proofread pass must be dispatched via runSubagent / Task tool. Do not role-play multiple reviewers within the same agent session.
Example (Review 1):
runSubagent(
agentName: "content-first-reviewer",
prompt: """
You are the first-pass reviewer. Below is the article draft:
[paste full article text]
Complete the Review 1 checklist and return your report
in the 'Review 1 Result' format.
"""
)
Each subagent prompt must clearly state:
- Its role and responsibilities for this pass
- The previous round's findings (if any)
- The full current article text
- The required output format
Revision Loop
If any round returns "Revise", follow this loop:
Revise → author edits → re-submit to same round (not from the top)
(if edits are major, restart from Review 1)
Scope judgement:
- Changes < 20% of content → re-run from the round that returned Revise
- Changes ≥ 20% of structure → restart from Review 1
Quick Reference
| Stage | Subagent Name | Focus |
|---|
| Review 1 | content-first-reviewer | Structure, completeness, basic quality |
| Review 2 | content-deep-reviewer | Logic, accuracy, depth, risks |
| Review 3 | content-chief-reviewer | Big-picture judgment, publish verdict |
| Proof 1 | proofreader-first | Line-by-line error scan |
| Proof 2 | proofreader-second | Verify fixes, catch new errors |
| Proof 3 | proofreader-final | Author cold-read, final sign-off |