campaign-chapter-review

name	campaign-chapter-review
description	Use when reviewing or editing D&D campaign narrative chapter markdown generated from session transcripts, especially when the user provides a campaign root path and a current summary/chapter file. Loads per-campaign canon docs, party state, planning pressure, character voice files, and examples; edits to a new sibling file unless explicitly told otherwise.

Campaign Chapter Review

Overview

Review campaign narrative chapters as canon-bearing fiction, not generic prose. The goal is a clean edited copy that preserves session truth, character voice, campaign continuity, and the user's established chapter style.

The user normally provides:

• a campaign root path, such as /home/kroussos/campaigns/Phandalin or a WSL UNC path to it
• a current summary/chapter markdown file, usually under summaries/YYYYMMDD/

Expected Campaign Layout

Given CAMPAIGN_ROOT, assume these files and folders unless the user says otherwise:

• CAMPAIGN_ROOT/docs/campaign_state.md — completed-content canon; do not contradict or replay
• CAMPAIGN_ROOT/docs/world_state.md — current setting/state truth
• CAMPAIGN_ROOT/docs/party.md — character state, bonds, items, active tensions
• CAMPAIGN_ROOT/docs/planning.md — forward pressure and GM intent; not necessarily revealed truth
• CAMPAIGN_ROOT/voice/ — narrator voice guides, typically {character}_voice.md
• CAMPAIGN_ROOT/examples/ — sample prose used to calibrate narrator/style

If a path is a Windows WSL UNC path like \\wsl.localhost\Ubuntu-24.04\home\..., read it directly or convert it to the matching WSL path for shell commands. Do not copy campaign files into the workspace unless the user asks.

Workflow

1. Read the current chapter structure first:

   • line count
   • headings
   • first section
   • remaining sections as needed

2. Load the canon bundle:

   • campaign_state.md
   • world_state.md
   • party.md
   • planning.md

3. Load only the voice/example files needed for narrators present in the chapter. Prefer headings and filenames to identify narrators.

4. Review in four passes:

   • Canon continuity: names, locations, timeline, item ownership, resolved/open threads
   • Narrator voice: each POV should match its voice file and examples, not generic polished prose
   • Transcript artifacts: garbled phrases, accidental player/table wording, malformed quotes, repeated generated motifs
   • Chapter clarity: scene transitions, causality, spatial orientation, whether facts/inferences/beliefs are distinguished

5. Write an edited copy as a sibling of the original file unless the user specifies a destination. Use a conservative suffix such as:

   • *-edited.md
   • *-reviewed.md
   • *-canon-edit.md

6. Verify with a diff and a small targeted scan for known risk strings.

Editing Rules

• Preserve the original file.
• Preserve event order unless a passage is clearly out of order and the session/chapter context proves the correction.
• Prefer small repairs over rewriting whole scenes.
• Do not invent session events, dialogue, or discoveries.
• Treat planning.md as hidden GM pressure unless the chapter text or canon docs show the party knows it.
• Keep character interiority character-specific:
  • factual uncertainty is fine in narration
  • revealed canon should not become stronger than the session supports
  • GM-only causes should remain implications unless already revealed
• Keep Markdown structure intact.

Review Lens

Use these labels mentally while editing:

• clarity: confusing wording, unclear referents, scene geometry, missing setup
• canon: contradiction with campaign/world/party state
• voice: narrator does not sound like their guide/example
• timeline: location/order drift, premature references to later events
• overclaim: text asserts something as fact that should remain belief or inference
• artifact: transcript cleanup issue, malformed phrase, duplicated model pattern

Preflight Script

For cheap mechanical checks, run:

python scripts/preflight_chapter.py CHAPTER.md

The script flags likely transcript artifacts and high-risk continuity words. It is not authoritative; use it as a checklist before and after editing.

Output

When finished, report:

• edited file path
• original file left untouched
• the highest-signal categories of edits made
• any residual concerns that need GM judgment