← Files RAG & GenAI CopilotARCHIVED FILE

references/orchestration.md

894 Bytes · Nov 11, 2026 · 12:49 UTC

↓ Download file

See the change to this file →

# Multi-agent execution policy
A skills-only plugin does not start background agents by itself. Only use independent agents when supported by current host or an authenticated backend. Do not fabricate role outputs.

Potential roles: Lead/architect, corpus analyst, retrieval engineer, security reviewer, evaluation engineer, implementation engineer, ops reviewer. Use only those required by task.

Plan task graph; parallelize independent read-only research and nonoverlapping edits. Freeze contracts and source/fixture versions. Reviewers inspect actual artifacts and do not self-certify implementation. Lead merges conclusions and records unresolved findings.

Each role returns scope, evidence/paths, commands executed, results, risks, approvals and follow-up. Bound concurrency, tokens, time and spend. Never deploy, purge an index or access privileged data without user consent and tools.

SHA-256: c3fb6cc5327338a062fb4ce2bd1eb85a96936bd3a5d524ee027762cab2eeb841