Skip to content

Latest commit

 

History

History
115 lines (87 loc) · 3.3 KB

WORKFLOW.md

File metadata and controls

115 lines (87 loc) · 3.3 KB

WORKFLOW

Step-by-step guide to running autonovel.

For the full technical pipeline specification, see PIPELINE.md.


Quick Start

# 1. Setup
cd ~/autonovel
cp .env.example .env   # Add your Anthropic API key

# 2. Generate a seed concept (or write your own in seed.txt)
uv run python seed.py

# 3. Create a branch for your novel
git checkout -b autonovel/my-novel

# 4. Run the full pipeline
uv run python run_pipeline.py --from-scratch

The pipeline will:

  1. Build the world, characters, outline, and voice (Phase 1)
  2. Draft all chapters sequentially (Phase 2)
  3. Revise through automated cycles + Opus review (Phase 3)
  4. Export to manuscript, PDF, ePub (Phase 4)

Running Phases Individually

# Foundation only
uv run python run_pipeline.py --phase foundation

# Drafting only
uv run python run_pipeline.py --phase drafting

# Revision only (with max cycle limit)
uv run python run_pipeline.py --phase revision --max-cycles 5

# Export only
uv run python run_pipeline.py --phase export

Manual Tools

Evaluation

uv run python evaluate.py --phase=foundation   # Score planning docs
uv run python evaluate.py --chapter=5           # Score a chapter
uv run python evaluate.py --full                # Score the whole novel

Revision

uv run python adversarial_edit.py all           # Find cuts in all chapters
uv run python apply_cuts.py all --types OVER-EXPLAIN REDUNDANT
uv run python reader_panel.py                   # 4-persona evaluation
uv run python review.py                         # Opus dual-persona review
uv run python gen_brief.py --auto               # Auto-generate revision brief
uv run python gen_revision.py 5 briefs/ch05.md  # Rewrite chapter from brief

Art (requires FAL_KEY)

uv run python gen_art.py style                  # Derive visual style
uv run python gen_art.py curate cover --n=6     # Generate cover variants
uv run python gen_art.py pick cover 3           # Select variant #3
uv run python gen_art.py ornaments-all          # Generate chapter ornaments
uv run python gen_art.py vectorize              # Convert to SVG → PDF
uv run python gen_cover_print.py art/cover.png --canvas-width 11.889 --canvas-height 8.75 --spine-width 0.639

Audiobook (requires ELEVENLABS_API_KEY)

uv run python gen_audiobook_script.py           # Parse all chapters
uv run python gen_audiobook.py --list-voices    # Browse voices
uv run python gen_audiobook.py --test 1         # Test chapter 1
uv run python gen_audiobook.py                  # Generate all
uv run python gen_audiobook.py --assemble       # Concatenate

Export

uv run python build_outline.py                  # Rebuild outline
uv run python build_arc_summary.py              # Rebuild summaries
python3 typeset/build_tex.py && cd typeset && tectonic novel.tex  # PDF

The Three Loops

INNER LOOP (agent, runs overnight):
  modify → evaluate → keep/discard → repeat

OUTER LOOP (you, when you check in):
  read results → steer program.md / evaluate.py / layer files
  → let the agent run again

REVIEW LOOP (after automated revision):
  send to Opus → parse review → fix top items → repeat
  → stop when no major unqualified items remain

You're not writing the novel. You're programming the system that writes the novel.