AEO Init

Source: github.com/psyduckler/aeo-skills Part of: AEO Skills Suite Output schema: aeo-config-v1

The workspace bootstrapper. One command, one config file, then aeo-baseline knows what to measure.

Requirements

• Python 3.9+
• No API keys (this skill never calls a provider)
• Optional: pip install jsonschema for full schema validation at write time

Usage

# Interactive — walks through every field
python3 scripts/init.py

# Flag-driven — best for agent use
python3 scripts/init.py \
  --brand Tabiji --domain tabiji.ai \
  --alias tabiji \
  --competitor tripadvisor.com --competitor lonelyplanet.com \
  --prompt "best travel planning tools" \
  --prompt "AI travel apps for itineraries" \
  --prompt-intent commercial --prompt-intent commercial

# Print the would-be config without writing
python3 scripts/init.py --brand X --domain x.com --prompt "best X" --dry-run

# Overwrite an existing config (safety: refuses unless --force)
python3 scripts/init.py --brand NewName --domain new.com --force

# Custom output path
python3 scripts/init.py --output /path/to/aeo.config.json ...

Interactive mode triggers automatically when neither --brand nor a useful flag combination is provided. Pass --interactive to force it even with flags.

Options

What gets written

A JSON file conforming to aeo-config-v1.json. The default skeleton includes:

• One Gemini provider (model gemini-3-flash-preview, env var GEMINI_API_KEY, grounding enabled)
• Sampling: 20 runs default, 5 concurrency, 3 retries per sample
• Limits: $25/day cap, $5 confirm-over threshold, 100 runs per campaign
• Scoring: aeo-v1 methodology
• Empty prompts list if none were provided (add later with --force or edit the file)

Domain normalization

Competitor and brand domains are normalized before writing:

• Lowercased
• https:// / http:// prefixes stripped
• Literal www. prefix stripped (using startswith, not the lstrip('www.') footgun that would corrupt world.com)
• Trailing / stripped

So https://WWW.TripAdvisor.com/ becomes tripadvisor.com.

Prompt IDs

When you pass --prompt "Best CRM for small business", the prompt is auto-assigned prompt_id = "best_crm_for_small_business". IDs are snake_case, deduplicated by truncation to 64 chars. Override by editing the JSON post-hoc.

Validation

After building the config, two validation layers run:

1. Built-in: required fields, api_key_env shape (^[A-Z][A-Z0-9_]*$), prompt_id shape (^[a-z0-9_]+$), intent enum membership.
2. JSON Schema (if jsonschema is importable): full structural validation against aeo-config-v1.json.

If validation fails, the script prints the errors and exits with a non-zero code without writing.

Pairs With

• aeo-baseline — reads the config this skill writes
• aeo-track — schedules aeo-baseline runs from this config
• Future prompt packs (e.g. aeo-pack-b2b-saas) will populate the prompts array

Notes

• This skill never calls a provider API and never reads secrets.
• The output is editable JSON — re-running with --force regenerates; editing in place is also fine.
• For a starter prompt list, run the (future) aeo-prompt-research-free skill first and feed its suggestions back into init via --prompt flags.