UNPKG

@kontextmind/kxm

Version:

KXM local-first multi-agent orchestration and operator dashboard

41 lines (40 loc) • 2.14 kB
[ { "id": "provenance-review", "source": "generic", "project": "provenance-demo", "target": "coordinator", "secretEnv": "KXM_WORKFLOW_SECRET", "delivery": "followUp", "promptTemplate": "Review {{task.id}}: {{task.summary}}. Collect independent peer findings, preserve contradictions, and make an evidence-backed decision.", "stages": [ { "id": "review", "label": "Independent review", "instructions": "Ask both eligible reviewers independently with one kxm_fanout call. Set correlationId to the exact run ID, timeoutMs to 120000, and idempotencyKeyPrefix to the deterministic value provenance-review:<runId>:review:<attempt>. The tool call itself must include workflowContext for this run, stage review, requirement independent peer reviews, and the current attempt; stating that context in a plan is not sufficient. Treat every returned messageId as the durable handle immediately. Use kxm_get to verify each stored message has the exact run, stage, requirement, and attempt binding. For every pending reply, call kxm_await with that messageId and timeoutMs 60000 (its maximum), repeating the call while it stays pending; never replace pending work with a new idempotency prefix. On interruption or uncertain delivery, repeat the exact fanout parameters so the hub returns the same messages. Re-read and verify each durable message after it replies. Compare findings, record contradictions and the chosen resolution, then cite only the bound durable replied message IDs.", "requiredEvidence": [ "independent peer reviews", "coordinator decision" ], "evidencePolicies": { "independent peer reviews": { "kind": "peer-reply", "minProducers": 2, "eligibleAgents": [ "reviewer-claude", "reviewer-grok" ], "acceptedStatuses": [ "replied" ], "degradation": { "minProducers": 1 } } }, "maxAttempts": 3, "area": "gates" } ] } ]