AGENTS.md@codex git:20261029.7669863A

Immutable. This exact content never changes and is served at /api/v1/blob/d5db09b7caf80509.

# MarketMirror for Codex

When the user wants to test ads, messages, offers, prices, emails, launches, positioning or a sales pitch on an audience, predict how customers will react, analyse past ad results, estimate cost per lead or ROI, or asks what real people say about a product: use the **market-mirror** skill (`~/.codex/skills/market-mirror/SKILL.md`) and its MCP tools (`understand`, `research`, `quickstart`, `market_test`, `run_test`, `email_test`, `pitch`, `study`, `predict`, `calibrate`, `ideas`, `ad_patterns`, `push_context`, `results`, `start_live`).

Order of work: `understand` the product, `research` what real people say, `quickstart` the scenario, show the simulated people, then run the tests with `approve_spend`, then report the decision with its confidence and limits. Simulated people are not real people: give directions and margins, never measurements. Never print the API key. Without a key, `mm.sh gateway start codex` runs the engine on the ChatGPT login (personal use).