Editing

Editing Skill

This skill provides comprehensive guidance for autonomous editing of manuscripts, covering multiple editing passes, consistency checks, and quality assurance.

Editing Philosophy

Editing for autonomous book writing requires systematic, database-backed verification rather than subjective judgment. Every editing pass should query the corpus to enforce consistency against established facts, voice, and timeline.

Multi-Pass Editing Strategy

Pass 1: Structural Consistency

Focus: Large-scale issues that affect narrative or argument coherence.

Checks:

• Timeline consistency (events in correct order)
• Character state tracking (no dead characters appearing alive)
• Location introduction (all locations described before use)
• Plot thread resolution (no abandoned arcs)
• Argument flow (nonfiction logical progression)

Queries: See consistency-check.surql for database queries.

Pass 2: Voice and Style

Focus: Maintaining consistent voice and improving prose quality.

Checks:

• Voice consistency against BRD voice profile
• Sentence variety and rhythm
• Word choice and precision
• Paragraph flow and transitions
• Dialogue authenticity

Guidance: See style.md for line-level editing techniques.

Pass 3: Citation and Grounding

Focus: Ensuring all factual claims are properly sourced.

Checks:

• Every factual claim has source citation
• Sources are reliable (per research skill criteria)
• Citations formatted correctly
• No hallucinated facts
• Primary sources used where appropriate

Queries:

-- Find sections with factual claims but no citations
SELECT * FROM section
WHERE content ~ 'factual_pattern'
AND count(->cites->source) = 0;

Pass 4: Continuity

Focus: Micro-level consistency across the manuscript.

Checks:

• Character details (name spelling, descriptions, traits)
• World-building details (location features, rules, systems)
• Timeline details (dates, durations, sequences)
• Previously established facts

Guidance: See continuity.md for detailed continuity checking.

Editing Workflow

Step 1: Load Context

Before editing any section, gather full context:

-- Get section content
SELECT * FROM section WHERE id = $section_id;

-- Get BRD voice profile
SELECT * FROM brd ORDER BY version DESC LIMIT 1;

-- Get recent sections for voice calibration
SELECT content FROM section
WHERE chapter = $chapter_id
ORDER BY sequence DESC
LIMIT 3;

Step 2: Run Consistency Checks

Execute automated consistency queries from consistency-check.surql:

• Character state violations
• Location introduction issues
• Timeline contradictions
• Uncited factual claims

Step 3: Generate Edit Report

For each issue found, create structured report:

## Edit Report: [Section ID]

### Structural Issues
- [ ] Character:anna appears in section 47 but status='dead' since section 32
- [ ] Location:safehouse used but not introduced

### Voice Issues
- [ ] Sentence length avg 8 words (BRD target: 15-20)
- [ ] Formal vocabulary in conversational narrative

### Citation Issues
- [ ] "SOE training took 6 weeks" - uncited factual claim
- [ ] Source reliability: 2 low-reliability sources used

### Continuity Issues
- [ ] Character description: "blue eyes" conflicts with section 12 "brown eyes"
- [ ] Timeline: Event dated "August 1943" conflicts with chapter 2 "September 1942"

Step 4: Apply or Suggest Fixes

Depending on mode:

• Report mode: Generate report for human review
• Auto-fix mode: Apply fixes that are unambiguous (spelling, formatting, citations)
• Interactive mode: Suggest fixes and wait for approval

Step 5: Verify Fixes

After applying fixes, re-run consistency checks to ensure issues resolved.

Voice Consistency Framework

Voice must match the BRD voice profile across all sections.

Voice Parameters to Track

From BRD voice profile:

• POV: First, third limited, omniscient
• Tense: Past, present, future
• Tone: Academic, conversational, literary, journalistic
• Formality: Casual, neutral, formal
• Sentence length: Average words per sentence
• Vocabulary: Simple, moderate, complex
• Metaphor frequency: Rare, moderate, frequent
• Dialogue ratio: Percentage of prose that is dialogue

Voice Verification Queries

-- Get voice profile from BRD
SELECT
    narrative_pov,
    tense,
    overall_tone,
    formality_level,
    avg_sentence_length,
    vocabulary_level,
    metaphor_frequency,
    dialogue_ratio
FROM brd
ORDER BY version DESC
LIMIT 1;

-- Compare section metrics to profile
-- (requires text analysis, typically LLM-based)

Voice Calibration

For each section, analyze:

1. Extract 200-word sample
2. Compare against BRD voice profile
3. Identify deviations (e.g., "too formal", "sentences too short")
4. Suggest rewrite or flag for human review

Timeline Verification

Timeline consistency is critical for both fiction and nonfiction.

Timeline Queries

-- Get all events in chronological order
SELECT * FROM event ORDER BY sequence ASC;

-- Find timeline contradictions
SELECT e1.name AS earlier_event, e2.name AS later_event
FROM event e1, event e2
WHERE e1.sequence > e2.sequence
AND e1.date < e2.date;

-- Verify character availability
SELECT
    section.id AS section_id,
    character.name AS character_name,
    character.status
FROM section->appears_in->character
WHERE character.status = 'dead'
AND section.sequence > (
    SELECT sequence FROM event
    WHERE name LIKE '%death of ' + character.name
    LIMIT 1
);

Timeline Fixes

When timeline issues found:

1. Minor discrepancy (hours/days): Adjust less-critical event
2. Major discrepancy (months/years): Flag for human review
3. Character state: Update character status or remove from scene

Citation Coverage

Every factual claim must be traceable to a source.

Citation Identification

Factual claims include:

• Dates, statistics, quotations
• Historical events, scientific findings
• Expert opinions, research results
• Specific details (measurements, names, places)

Uncited Claim Detection

-- Find sections with no citations (potential issue)
SELECT id, content FROM section
WHERE count(->cites->source) = 0;

-- Find low citation density
SELECT
    id,
    word_count,
    count(->cites->source) AS citation_count,
    word_count / count(->cites->source) AS words_per_citation
FROM section
WHERE word_count / count(->cites->source) > 500;

Citation Quality

Verify citations meet standards:

• Source reliability ≥ medium
• Citation specific (page numbers, timestamps)
• Primary sources for key claims
• Multiple sources for controversial claims

Continuity Checking

See continuity.md for detailed continuity verification:

• Character details consistency
• World-building elements
• Established facts
• Timeline coherence

Edit Decision Framework

Quality Thresholds

Before marking a section as "edited":

• Zero critical consistency issues
• All factual claims cited
• Voice matches BRD profile
• Timeline verified
• Continuity checked

Output Formats

Edit Report

# Edit Report: Chapter 3, Section 2

**Status**: Needs revision
**Issues found**: 7 (2 critical, 3 high, 2 medium)

## Critical Issues
1. Character state violation: Anna appears but marked dead
2. Timeline contradiction: Event dated after subsequent event

## High Priority Issues
3. Uncited claim: "SOE training took 6 weeks"
4. Uncited claim: "Lyon network had 47 members"
5. Source reliability: Wikipedia cited for key historical fact

## Medium Priority Issues
6. Voice: Sentences avg 8 words (target: 15-20)
7. Continuity: Anna's eye color changed from brown to blue

## Recommended Actions
- Review timeline in sections 32-47
- Add citations for historical claims
- Revise for sentence variety
- Verify character descriptions against database

Auto-Fix Summary

# Auto-Fix Summary: Chapter 3, Section 2

**Changes applied**: 5

1. Fixed spelling: "occured" → "occurred"
2. Removed character:anna from section (status=dead)
3. Updated location:safehouse.introduced = true
4. Added citation: source:soe_manual_1942
5. Standardized date format: "Aug 1942" → "1942-08-15"

**Verification**: All critical issues resolved

Supporting Files

• continuity.md — Detailed continuity checking strategies
• style.md — Line-level editing techniques and guidelines
• consistency-check.surql — Database queries for automated checks

Integration with Writer Agent

The editing skill should be invoked after writing completes:

1. Writer completes section
2. Editor runs automated checks
3. Issues flagged or auto-fixed
4. Section marked "edited" or "needs revision"
5. If needs revision, writer re-engages with edit report

Best Practices

1. Database-driven: Always verify against database, not memory
2. Systematic: Run all passes in order
3. Document changes: Log all auto-fixes for transparency
4. Flag ambiguity: When unclear, flag for human review
5. Preserve voice: Don't over-edit to remove author's style
6. Verify fixes: Re-run checks after applying changes