A practical Python repo that turns the X article strategy into a reusable workflow for building Fortnite/UEFN map ideas at scale.

• Concept generation prompts for high-retention map ideas
• Winner filtering based on session length, monetization potential, complexity, and asset count
• Verse scaffold prompts for moving top concepts into implementation
• Retention analysis prompts for fixing weak session loops
• Metadata/SEO prompts for Fortnite internal discovery
• Analytics diagnosis prompts for publish -> analyze -> iterate loops
• Portfolio math for rough payout estimation using the engagement-pool model from the article

src/fortnite_ai_map_pipeline/
  models.py
  scoring.py
  prompts.py
  pipeline.py
  client.py
  cli.py
examples/sample_concepts.json
tests/
docs/plans/

python -m venv .venv
source .venv/bin/activate
pip install -e .[dev]

For live Anthropic-backed generation:

pip install -e .[dev,ai]
cp .env.example .env
# fill in ANTHROPIC_API_KEY

Render the concept prompt:

PYTHONPATH=src python -m fortnite_ai_map_pipeline.cli prompt-concepts --genre social --count 12

Summarize a sample batch:

PYTHONPATH=src python -m fortnite_ai_map_pipeline.cli summarize examples/sample_concepts.json

Render a retention prompt from the first sample concept:

PYTHONPATH=src python -m fortnite_ai_map_pipeline.cli render-prompt examples/sample_concepts.json --kind retention

Run live generation (requires optional dependency + API key):

PYTHONPATH=src python -m fortnite_ai_map_pipeline.cli generate --genre survival --count 20

prompts.build_concept_generation_prompt() encodes the article's constraints:

• avoid generic map ideas
• optimize for retention and social mechanics
• target 15+ minute sessions
• prefer fast-build concepts over heavy Verse complexity

scoring.compute_viability_score() and filter_winners() rank concepts by:

• target session length
• monetization potential
• implementation speed
• asset count discipline

prompts.build_verse_scaffold_prompt() prepares the data needed for AI-assisted Verse generation.

prompts.build_retention_analysis_prompt() explicitly asks about:

• the 30-second hook
• the 3-minute wall
• the 10-minute reward
• social stickiness
• return motivation

prompts.build_metadata_prompt() captures map naming, descriptions, tags, and thumbnail direction.

prompts.build_analytics_diagnosis_prompt() supports the article's feedback loop:

• publish
• inspect drop-off
• diagnose with AI
• implement fast fixes
• republish

pytest -q

Current suite covers:

• concept scoring
• winner filtering
• payout estimate helper
• portfolio aggregation
• all prompt builders

A Codex-first Next.js web app now lives in web/.

• server-side generation via local codex exec
• live /api/health route
• live /api/generate route
• ranked concepts with viability scores and revenue estimates
• auto-built Verse / retention / metadata prompt kits per concept

cd web
pnpm install
pnpm dev

Production build:

cd web
pnpm build
pnpm start

The web app uses the existing local Codex auth/session on the machine. If ~/.codex/auth.json exists, the UI should show Codex auth as ready.

• The deterministic parts are testable offline.
• The live generation layer is intentionally optional so the repo still works without external API access.
• This is a starter kit for the workflow, not a full UEFN editor integration.