input-output-contract.md

Input And Output Contract

This repository uses a shared contract so GEO skills do not drift into incompatible prompt bundles.

Required Input Fields

Every GEO skill should normalize its user input into the following base structure:

skill_id: ""
brand_name: ""
brand_site: ""
brand_summary: ""
category: ""
region: ""
language: ""
target_audience: ""
primary_goal: ""
freshness_window_days: 30
source_policy:
  require_official_sources: true
  require_current_web_verification: true
  allow_forums: false
competitor_seeds: []
must_cover_queries: []
must_avoid_queries: []
deliverable_config:
  output_dir: ""
  article_count: 0
  include_html_index: true
notes: ""

Required Base Inputs

The following fields are required for every factual GEO skill:

• brand_name
• brand_site
• category
• region
• language
• primary_goal

Input Rules

• Default values must be visible in the template, not hidden in prompt prose.
• Freshness requirements must be explicit.
• Source policy must be explicit.
• Output configuration must be explicit.
• If a skill needs extra fields, extend the base contract instead of replacing it.

Output Layout

Every run should write to a stable directory:

outputs/<skill-id>/<run-date>-<brand-slug>/
├── manifest.json
├── input-brief.json
├── research-summary.md
├── sources.json
├── quality-report.json
├── deliverables/
└── index.html

Required Output Files

• manifest.json: run metadata and versioning
• input-brief.json: normalized input snapshot
• research-summary.md: operator-readable summary
• sources.json: citation ledger
• quality-report.json: validation results and open issues

Output Rules

• Any factual comparison must be traceable to a source entry.
• Any dated claim must carry freshness evidence.
• Deliverables should answer directly before adding long background.
• If verification fails, the output should record the blocker instead of fabricating certainty.