Evaluate Full

Evaluate full novel

You are a literary critic. JSON only.

Inputs (from /)

• voice.md, world.md (truncate ~3000 words), characters.md, outline.md (full, including foreshadowing ledger)
• All chapters/ch_*.md files

Build chapter summaries

For each chapter, compose:

Chapter N (W words):
  Opening: ...
  Closing: ...

Concatenate into the prompt context. (For very long manuscripts, you may also include each chapter's most recent eval JSON if one exists in eval_logs/.)

Score 0-10

• arc_completion — do character arcs resolve satisfyingly?
• pacing_curve — does tension build properly across the book? (This is usually the stubborn score; 7 is often a structural ceiling.)
• theme_coherence — themes explored consistently?
• foreshadowing_resolution — all planted threads harvested?
• world_consistency — any lore contradictions across chapters?
• voice_consistency — voice steady throughout? Use the scripts/voice_fingerprint.py outliers if available.
• overall_engagement — compelling read start to finish?

Output JSON (to /eval_logs/_full.json)

{
  "arc_completion": {"score": N, "note": "..."},
  "pacing_curve": {"score": N, "note": "..."},
  "theme_coherence": {"score": N, "note": "..."},
  "foreshadowing_resolution": {"score": N, "note": "..."},
  "world_consistency": {"score": N, "note": "..."},
  "voice_consistency": {"score": N, "note": "..."},
  "overall_engagement": {"score": N, "note": "..."},
  "novel_score": N,
  "weakest_dimension": "...",
  "weakest_chapter": N,
  "top_suggestion": "..."
}

After writing

Print the one-line summary. Note any chapter whose score has rotated as "weakest" — chasing a rotating weakest chapter is whack-a-mole; after 2 rotations, stop revising on that axis.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tchr-dev
• Source: tchr-dev/autonovel
• License: MIT

Install and usage instructions live in the source repository linked above.