Editorial Review Loop

Apply human editorial corrections to a Phase 4 newsletter and produce an updated version.

Quick Start

1. •Read the corrections file at workspace/YYYY-MM_editorial_corrections.md
2. •Read the current newsletter at output/YYYY-MM_month_newsletter.md
3. •Read reference/editorial-intelligence.md for calibrated rules
4. •Apply each correction to the newsletter
5. •Update intelligence files with new patterns
6. •Validate the updated newsletter
7. •Capture learnings in LEARNINGS.md

Inputs

• •Corrections File: workspace/YYYY-MM_editorial_corrections.md (required)
• •Current Newsletter: output/YYYY-MM_month_newsletter.md (required)
• •Editorial Intelligence: reference/editorial-intelligence.md (context)
• •Source Intelligence: reference/source-intelligence/meta-analysis.md (context)

Output

• •Updated Newsletter: output/YYYY-MM_month_newsletter.md (overwrite in place)
• •Updated Intelligence: Changes to reference/editorial-intelligence.md if corrections reveal new patterns
• •New Learnings: Appended to LEARNINGS.md

Corrections File Format

The human writes corrections to workspace/YYYY-MM_editorial_corrections.md. The required structure, sections, and per-item table format are defined in the single authoritative spec: correction-format.md. Use that document as the source of truth for how to structure and label all corrections.

When running this skill, assume the corrections file strictly follows references/correction-format.md and treat any deviations as format errors to be surfaced back to the human editor.

• •Item name, what to do with it (remove / compress / bundle)

Bundling Corrections

[Items that should be combined or separated] | Bundle | Items | Treatment |

Depth Corrections

[Items that need more or less detail]

• •Item name: expand / compress, what detail to add/remove

Tone/Framing Corrections

[Narrative or framing changes needed]

## Core Workflow

### Step 1: Parse Corrections

Read the corrections file and categorize each correction:

| Type | Action |
|------|--------|
| **Theme correction** | Rewrite lead section with new theme |
| **Missing item** | Add item with source URL and appropriate detail |
| **Wrong item** | Remove, compress, or bundle as specified |
| **Bundling correction** | Combine items or split them per instructions |
| **Depth correction** | Expand or compress specific items |
| **Tone/framing** | Rewrite affected sections |

### Step 2: Apply Corrections to Newsletter

For each correction, modify the newsletter in place:

1. **Theme changes**: Replace the lead section title, framing paragraph, and reorganize lead items
2. **Additions**: Insert new items in the appropriate section, following the content-format-spec
3. **Removals**: Delete the item entirely or compress into a bundle
4. **Bundling**: Merge items into a single bullet with sub-items, or split a bundle into separate bullets
5. **Depth changes**: Expand with sub-bullets or compress to a single line
6. **Tone changes**: Rewrite the affected paragraph(s)

### Step 3: Validate Updated Newsletter

Run validation:
```bash
bash .github/skills/newsletter-validation/scripts/validate_newsletter.sh output/YYYY-MM_month_newsletter.md

Check:

• • All corrections applied (count corrections vs changes made)
• • Validation passes (0 errors)
• • Line count in expected range (100-170 lines)
• • No new forbidden patterns introduced
• • All new items have source URLs

Step 4: Update Intelligence Files

If corrections reveal new editorial patterns, update:

1. •reference/editorial-intelligence.md: Add new theme triggers, adjust selection weights, add expansion/compression rules
2. •reference/source-intelligence/meta-analysis.md: Note new source gaps or patterns
3. •Selection criteria: If the correction shows a weight was miscalibrated

Only update when the correction reveals a PATTERN, not a one-off preference.

Step 5: Capture Learnings

Append to LEARNINGS.md:

• •One learning per correction type that reveals a generalizable pattern
• •Include: ID, Lesson, Evidence (the specific correction), Fix (what to change in the system)

Step 6: Score V1 vs V2

If a scoring rubric exists for this month (tools/score-v2-rubric.sh), run it on the updated newsletter and compare to the pre-correction score.

Report:

• •V1 score (pre-correction)
• •V2 score (post-correction)
• •Dimensions that improved
• •Rework cycles used

Reference

• •Correction Format - Standard correction file format and examples
• •Editorial Intelligence - Theme detection, selection weights, treatment patterns
• •Source Intelligence - Cross-cycle findings

Done When

• • All corrections from the corrections file have been applied
• • Updated newsletter passes validate_newsletter.sh
• • Intelligence files updated (if patterns detected)
• • Learnings captured in LEARNINGS.md
• • V1 vs V2 comparison reported