Eval Loop — Orchestrator

Meta process skill. Turns a measurable initiative into a domain-scoped loop workspace where strategy artifacts, execution artifacts, eval snapshots, result rows, and promoted learnings live together.

Core Question: "Can future agents improve this measurable surface by reading one loop folder instead of reconstructing history from scattered skill outputs?"

Critical Gates

1. Measurable surface required. If the user cannot name a page, campaign, post series, ad set, email sequence, outreach motion, or other observable surface, return NEEDS_CONTEXT and recommend discover.
2. Metric path required. The loop must name at least one primary metric and where it will come from, even if the baseline is not known yet. No metric path -> no loop.
3. No skill-centered folders. Do not create .agents/skill-artifacts/{skill-name}/.... Eval loops are organized by measurable initiative.
4. Execution boundary. This stack may execute marketing/content assets. It does not deploy code, publish to platforms, build app UI, or mutate external systems.
5. No unattended infinite marketing loops. Borrow autoresearch's ledger and keep/discard discipline, not its "run forever" posture. Human approval gates publishing and live-surface changes.

Reference

Read before writing or modifying any loop artifact: