Contract
- Input: task description requiring multiple agents, available tools, quality requirements, latency constraints.
- Output: orchestration architecture + capability matrix + handoff contracts + evaluation rubric.
- Scope: designs orchestration patterns; does not execute multi-agent workflows unless explicitly instructed.
- Rule: designs orchestration patterns; does not execute multi-agent workflows unless explicitly instructed.
- Rule: documented standards override defaults; explicit project rules take precedence.
- Rule: if blocked by missing context or dependencies, surface the blocker before proceeding.
Provenance
| Question |
Answer |
| What is the source of truth? |
The user's request, originating spec/issue, and the skill's declared outputs |
| What is in scope? |
Work covered by the skill's acceptance criteria and completion rules |
| What is explicitly out of scope? |
Files, behaviors, and decisions outside the skill's declared boundary |
| Who or what consumes this artifact afterward? |
The next skill in the workflow or the user |
| What evidence proves it is done? |
Completion criteria met, artifact saved, validation passed |
| What risk remains? |
Subjective judgment calls, missing context, or external dependency failures |
Artifact
Emit AgentOrchestrationArtifact as both:
- JSON:
.agents/skills/platform/artifacts/agent-orchestration/{identifier}.json
- Markdown view: same filename with
.md extension
Agent Orchestration
Design a multi-agent orchestration system with clear coordination boundaries, tool use contracts, memory architecture, and observability.
Process
1. Frame the orchestration problem
- Task decomposition: which parts of the workflow need separate agents?
- Agent count and types: supervisor, specialist workers, validator, synthesizer.
- Latency and throughput requirements: sequential vs parallel execution.
- Failure modes: agent timeout, tool failure, ambiguous output, hallucination.
Completion criterion: task decomposition documented; agent count and types defined; latency requirements stated.
2. Agent roles and tool use
- Role definition: name, responsibility, success criteria, failure handling.
- Tool allocation: which tools each agent can use; least-privilege principle.
- Tool contracts: input schema, output schema, retry policy, timeout, idempotency.
- Context scoping: what context each agent receives; minimisation to reduce noise and cost.
Completion criterion: capability matrix saved with roles, tools, and contracts.
3. Memory architecture
- Short-term memory: conversation history, scratchpad, working state within a single agent run.
- Long-term memory: knowledge graph, vector store, file system for cross-session persistence.
- Shared memory: state visible to multiple agents (board, queue, registry).
- Episodic memory: execution traces for debugging and improvement.
- Memory access policy: read/write permissions, retention, summarisation strategy.
Completion criterion: memory architecture documented with access policies.
4. Coordination patterns and handoffs
- Patterns: supervisor / worker, peer collaboration, pipeline / DAG, router / classifier.
- Handoff contracts: trigger conditions, payload schema, timeout, retry, fallback agent.
- Escalation paths: when to escalate to human or fallback agent; circuit breakers.
- Idempotency: how to handle duplicate or out-of-order messages.
Completion criterion: handoff contracts and escalation policy saved.
5. Evaluation and observability
- Coordination quality metrics: task completion rate, handoff success rate, agent latency distribution, tool call accuracy.
- End-to-end tracing: span per agent, per tool call, per handoff.
- Logging: structured logs with correlation IDs; redaction of sensitive data.
- Feedback loops: human-in-the-loop checkpoints, self-correction triggers, continuous improvement signals.
Completion criterion: evaluation rubric and observability plan saved.
Rules
- Rule: define explicit handoff contracts with schema, timeout, and fallback; never assume implicit coordination.
- Rule: apply least-privilege tool access per agent; document capability matrix.
- Rule: include circuit breakers and escalation paths for agent failures and ambiguous outputs.
- Rule: instrument every agent run with tracing, structured logging, and quality metrics.
- Rule: document memory access policies (read/write permissions, retention, summarisation) for each memory tier.
Completion
- the skill's completion criteria are explicitly checked
- the artifact is saved and validated
- any blockers or skipped validations are documented
- the next consumer is identified or the work is handed off
@include .agents/skills/platform/contract-base.xml
1---2name: "agent-orchestration"3category: "ai"4maturity: "stable"5version: "1"6description: "Design agent orchestration — multi-agent patterns, tool use, memory, handoffs, evaluation — with explicit coordination boundaries and observability."7license: "MIT"8capabilities:9 - apply agent orchestration methods within explicit ai boundaries10 - analyze design agent orchestration — multi-agent patterns, tool use, memory, handoffs, evaluation — with explicit coordination boundaries and observability with assumptions, constraints, and failure modes11 - produce and validate a ai artifact with evidence, decisions, and next steps12outputs: "orchestration architecture + capability matrix + handoff contracts + evaluation rubric."13sideEffects: []14dependencies: []15stopCondition: "Architecture diagram saved; capability matrix complete; handoff contracts defined; evaluation rubric present."16risk: "low"17trustTier: "1"18maxIterations: "6"19promptVersion: "2.0"20artifactType: "agent"21modelTier: "reasoning"22evaluators: ["behavioral", "regression"]23fixturesPath: ""24diataxis: "how-to"25tags: ["ai"]26compatibility: []27approvalRequired: "false"28approvalFor: []29metadata:30 source: https://skillmd.com/skills/quantumquirkxyz/agent-orchestration31---3233## Contract3435- Input: task description requiring multiple agents, available tools, quality requirements, latency constraints.36- Output: orchestration architecture + capability matrix + handoff contracts + evaluation rubric.37- Scope: designs orchestration patterns; does not execute multi-agent workflows unless explicitly instructed.38- Rule: designs orchestration patterns; does not execute multi-agent workflows unless explicitly instructed.39- Rule: documented standards override defaults; explicit project rules take precedence.40- Rule: if blocked by missing context or dependencies, surface the blocker before proceeding.4142## Provenance4344| Question | Answer |45|---|---|46| What is the source of truth? | The user's request, originating spec/issue, and the skill's declared outputs |47| What is in scope? | Work covered by the skill's acceptance criteria and completion rules |48| What is explicitly out of scope? | Files, behaviors, and decisions outside the skill's declared boundary |49| Who or what consumes this artifact afterward? | The next skill in the workflow or the user |50| What evidence proves it is done? | Completion criteria met, artifact saved, validation passed |51| What risk remains? | Subjective judgment calls, missing context, or external dependency failures525354## Artifact5556Emit `AgentOrchestrationArtifact` as both:57- JSON: `.agents/skills/platform/artifacts/agent-orchestration/{identifier}.json`58- Markdown view: same filename with `.md` extension596061# 62# 63# Agent Orchestration6465Design a **multi-agent orchestration** system with clear coordination boundaries, tool use contracts, memory architecture, and observability.6667## Process6869### 1. Frame the orchestration problem70- Task decomposition: which parts of the workflow need separate agents?71- Agent count and types: supervisor, specialist workers, validator, synthesizer.72- Latency and throughput requirements: sequential vs parallel execution.73- Failure modes: agent timeout, tool failure, ambiguous output, hallucination.7475**Completion criterion:** task decomposition documented; agent count and types defined; latency requirements stated.7677### 2. Agent roles and tool use78- Role definition: name, responsibility, success criteria, failure handling.79- Tool allocation: which tools each agent can use; least-privilege principle.80- Tool contracts: input schema, output schema, retry policy, timeout, idempotency.81- Context scoping: what context each agent receives; minimisation to reduce noise and cost.8283**Completion criterion:** capability matrix saved with roles, tools, and contracts.8485### 3. Memory architecture86- Short-term memory: conversation history, scratchpad, working state within a single agent run.87- Long-term memory: knowledge graph, vector store, file system for cross-session persistence.88- Shared memory: state visible to multiple agents (board, queue, registry).89- Episodic memory: execution traces for debugging and improvement.90- Memory access policy: read/write permissions, retention, summarisation strategy.9192**Completion criterion:** memory architecture documented with access policies.9394### 4. Coordination patterns and handoffs95- Patterns: supervisor / worker, peer collaboration, pipeline / DAG, router / classifier.96- Handoff contracts: trigger conditions, payload schema, timeout, retry, fallback agent.97- Escalation paths: when to escalate to human or fallback agent; circuit breakers.98- Idempotency: how to handle duplicate or out-of-order messages.99100**Completion criterion:** handoff contracts and escalation policy saved.101102### 5. Evaluation and observability103- Coordination quality metrics: task completion rate, handoff success rate, agent latency distribution, tool call accuracy.104- End-to-end tracing: span per agent, per tool call, per handoff.105- Logging: structured logs with correlation IDs; redaction of sensitive data.106- Feedback loops: human-in-the-loop checkpoints, self-correction triggers, continuous improvement signals.107108**Completion criterion:** evaluation rubric and observability plan saved.109110## Rules111112- Rule: define explicit handoff contracts with schema, timeout, and fallback; never assume implicit coordination.113- Rule: apply least-privilege tool access per agent; document capability matrix.114- Rule: include circuit breakers and escalation paths for agent failures and ambiguous outputs.115- Rule: instrument every agent run with tracing, structured logging, and quality metrics.116- Rule: document memory access policies (read/write permissions, retention, summarisation) for each memory tier.117118## Completion119120- the skill's completion criteria are explicitly checked121- the artifact is saved and validated122- any blockers or skipped validations are documented123- the next consumer is identified or the work is handed off124---125@include .agents/skills/platform/contract-base.xml