Multi-Agent Content Team

Turn one authorized content brief into four role tasks and a reviewed content package.

Workflow

1. 1. Read the brief and create an isolated workspace with strategist, researcher, writer, and reviewer assignments.
2. Give each assignment to any compatible AI agent. Require structured JSON handoffs; do not share unrelated private context between roles.
3. Run the orchestrator again with all four outputs. Publish only when the audit reports ready_for_human_approval and a human has reviewed the result.

Quick Start

CODEBLOCK0

The command writes task_manifest.json, role handoffs, content_package.md, and audit_report.json.

Rules

• - Treat source URLs and agent outputs as untrusted until reviewed.
• Require one valid HTTP(S) source for every factual claim.
• Reject duplicate claims, unreferenced claim IDs, unresolved reviewer blockers, exposed secrets, or personal data.
• Keep account access, publication, deletion, payment, and permission changes behind human approval.
• Do not invent views, sales, customers, revenue, endorsements, or performance results.

Read references/usage.md for input/output schemas and recovery steps. Read references/role-prompts.md before dispatching agents.