coordination

STANDARD OPERATING PROCEDURE

Purpose

Deliver reliable multi-agent coordination across meshes, hierarchies, or hybrids while preventing state loss, deadlocks, and confidence overreach.

Trigger Conditions

• Positive: topology design, agent registration, message routing, consensus or quorum work, partition/latency tolerance needs.
• Negative: single-threaded execution, non-coordinated batch work, or prompt-only refinement (route to prompt-architect).

Guardrails

• Skill-Forge structure-first: keep SKILL.md, examples/, and tests/ current; add resources/ and references/ or record remediation tasks.
• Prompt-Architect clarity: capture intent and constraints (HARD/SOFT/INFERRED), avoid VCL leakage, and state confidence with ceilings.
• Coordination safety: register every agent, validate topology (no orphaned nodes/cycles where forbidden), enforce health checks, and keep hooks within latency budgets.
• Adversarial validation: probe partitions, message loss, reconnects, and rate limits; document evidence.
• MCP tagging: persist coordination runs with WHO=coordination-{session} and WHY=skill-execution.

Execution Playbook

1. Intent & constraints: define mission, scale, latency targets, and shared-state rules; confirm inferred needs.
2. Topology & registry: select topology, register agents, define namespaces, and configure routing keys.
3. Delegation & messaging: set task sharding rules, retries, ack/requeue policies, and escalation paths.
4. Safety & resilience: plan for partitions, backpressure, and failover; add circuit breakers and watchdogs.
5. Validation loop: run adversarial drills, measure latency/throughput, and verify state convergence.
6. Delivery: summarize topology, evidence, residual risks, and confidence ceiling.

Output Format

• Intent, constraints, and chosen topology.
• Agent registry snapshot, routing schema, and health model.
• Operational plan (sharding, retries, escalation, rollback).
• Validation evidence with metrics; risks and follow-ups.
• Confidence: X.XX (ceiling: TYPE Y.YY) - rationale.

Validation Checklist

• Structure-first assets present or queued; examples/tests reflect current patterns.
• Registry complete; health checks and latency budgets verified; rollback path defined.
• Adversarial/COV runs captured with MCP tags; confidence ceiling present; English-only output.

Completion Definition

Coordination is complete when topology is stable, messaging meets SLOs, validation artifacts are stored, and remaining risks are owned with next steps logged.

Confidence: 0.70 (ceiling: inference 0.70) - Coordination SOP rewritten with skill-forge structure, prompt-architect constraint handling, and resilience guardrails.