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
- Intent & constraints: define mission, scale, latency targets, and shared-state rules; confirm inferred needs.
- Topology & registry: select topology, register agents, define namespaces, and configure routing keys.
- Delegation & messaging: set task sharding rules, retries, ack/requeue policies, and escalation paths.
- Safety & resilience: plan for partitions, backpressure, and failover; add circuit breakers and watchdogs.
- Validation loop: run adversarial drills, measure latency/throughput, and verify state convergence.
- 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.
1---2name: coordination3description: Coordinate distributed agents with resilient topologies, synchronized state, and evidence-backed communication patterns.4---5
6
7
8
9## STANDARD OPERATING PROCEDURE
10
11### Purpose
12Deliver reliable multi-agent coordination across meshes, hierarchies, or hybrids while preventing state loss, deadlocks, and confidence overreach.
13
14### Trigger Conditions
15- **Positive:** topology design, agent registration, message routing, consensus or quorum work, partition/latency tolerance needs.
16- **Negative:** single-threaded execution, non-coordinated batch work, or prompt-only refinement (route to prompt-architect).
17
18### Guardrails
19- **Skill-Forge structure-first:** keep `SKILL.md`, `examples/`, and `tests/` current; add `resources/` and `references/` or record remediation tasks.
20- **Prompt-Architect clarity:** capture intent and constraints (HARD/SOFT/INFERRED), avoid VCL leakage, and state confidence with ceilings.
21- **Coordination safety:** register every agent, validate topology (no orphaned nodes/cycles where forbidden), enforce health checks, and keep hooks within latency budgets.
22- **Adversarial validation:** probe partitions, message loss, reconnects, and rate limits; document evidence.
23- **MCP tagging:** persist coordination runs with WHO=`coordination-{session}` and WHY=`skill-execution`.
24
25### Execution Playbook
261. **Intent & constraints:** define mission, scale, latency targets, and shared-state rules; confirm inferred needs.
272. **Topology & registry:** select topology, register agents, define namespaces, and configure routing keys.
283. **Delegation & messaging:** set task sharding rules, retries, ack/requeue policies, and escalation paths.
294. **Safety & resilience:** plan for partitions, backpressure, and failover; add circuit breakers and watchdogs.
305. **Validation loop:** run adversarial drills, measure latency/throughput, and verify state convergence.
316. **Delivery:** summarize topology, evidence, residual risks, and confidence ceiling.
32
33### Output Format
34- Intent, constraints, and chosen topology.
35- Agent registry snapshot, routing schema, and health model.
36- Operational plan (sharding, retries, escalation, rollback).
37- Validation evidence with metrics; risks and follow-ups.
38- **Confidence:** `X.XX (ceiling: TYPE Y.YY) - rationale`.
39
40### Validation Checklist
41- Structure-first assets present or queued; examples/tests reflect current patterns.
42- Registry complete; health checks and latency budgets verified; rollback path defined.
43- Adversarial/COV runs captured with MCP tags; confidence ceiling present; English-only output.
44
45### Completion Definition
46Coordination is complete when topology is stable, messaging meets SLOs, validation artifacts are stored, and remaining risks are owned with next steps logged.
47
48Confidence: 0.70 (ceiling: inference 0.70) - Coordination SOP rewritten with skill-forge structure, prompt-architect constraint handling, and resilience guardrails.
majiayu000