← Files Product & App ArchitectARCHIVED FILE

references/multi-agent-orchestration.md

1.03 KB · Nov 20, 2026 · 06:43 UTC

↓ Download file

See the change to this file →

# Honest Specialist-Agent Coordination

An agent description is **not** an independently running agent. Real subagents require host-provided tool capability or authorized backend. Otherwise use sequential role-based passes and label them accurately.

Possible roles: product researcher, UX analyst, systems architect, security reviewer, feasibility engineer, editor/QA. Activate only what contributes independent evidence. For simple requests use a single workflow.

Lead owns problem statement, shared assumptions, acceptance tests, task dependencies, conflicts and final synthesis. Parallel research and architecture review are suitable; dependent UX->architecture work follows order. Reviewers read source/evidence independently and do not rubber-stamp implementation.

Each delegated task must state scope, data available, permission boundaries, expected artifact and completion evidence. Prevent duplicate work, contradictory assumptions and hidden writes. Never claim agent tools, repository edits, deployments or reviews without execution evidence.

SHA-256: 1bec85fd8c08e726aa29f739caa67165bbf01b414a0c38234ae073d1eb2f151e