| name | mkg-orchestrate |
| description | Run memory-grounded multiagent orchestration with Meta Knowledge Graph in Codex. Use when the user asks to execute a task through MKG, recall prior learnings first, route memory into subagents, verify phases, and capture durable learnings or decisions. |
MKG Orchestrate
Mirror the Claude Code /mkg-orchestrate command as closely as Codex allows.
Before acting, read the canonical command spec at
../../commands/mkg-orchestrate.md and follow it as the source of truth. Treat
this skill file as the Codex adapter layer, not a fork of the workflow.
Codex Adaptations
- Treat the user's explicit skill invocation text or trailing request as the
command's
$ARGUMENTS.
- Start with MKG recall. Prefer the
$mkg-recall skill when available; otherwise
call the mounted project_get_context or neo4j_read_cypher MCP tools
directly.
- Use Codex subagent or multi-agent tools when they are available. If this
session does not expose subagent tools, keep the same phase discipline in the
main session: recall first, plan, implement one phase at a time, verify with
commands, then capture only durable signal.
- Keep subagents read-only against MKG. The main orchestrator owns any
project_add_learning writes to avoid duplicate or
conflicting captures.
- When querying Neo4j through MCP, wrap temporal values with
toString(...) so
date/datetime values do not serialize as {}.
- Preserve the canonical command's anti-pattern checks, verification before
advancement, and over-capture guardrails.
Output
Return the memory brief, phase status, verification evidence, and any deliberate
MKG captures. Do not dump raw transcripts or secrets.
