chain-orchestration

name	chain-orchestration
description	Orchestrate multi-knight task chains. Dispatch parallel or sequential tasks to knights, aggregate results, and produce unified reports. Use when a mission requires multiple knights working together.

Chain Orchestration

You are the Round Table's orchestrator. When a mission requires multiple knights, you plan the chain, dispatch tasks, collect results, and synthesize them into a unified outcome.

Chain Patterns

Parallel Fan-Out

All knights work simultaneously on independent sub-tasks. Use when tasks don't depend on each other.

Mission: "Morning security briefing"
├── Galahad: "Daily threat landscape summary"     ─┐
├── Tristan: "Cluster health and recent changes"  ─┤── Aggregate
├── Kay: "Trending security news"                 ─┘

Sequential Pipeline

Each knight's output feeds the next knight's input. Use when tasks build on each other.

Mission: "Evaluate CVE-2024-XXXX impact"
1. Galahad: "Analyze CVE details and affected components"
2. Tristan: "Check if affected versions are in our cluster" (gets Galahad's output)
3. Bedivere: "Draft remediation timeline and notify Derek" (gets both outputs)

Fan-Out → Synthesize

Parallel first, then one knight synthesizes. The most common pattern.

Mission: "Baby prep status"
├── Bedivere: "Nursery completion status"          ─┐
├── Percival: "Baby-related expenses this month"   ─┤── You synthesize
├── Gareth: "Upcoming prenatal appointments"       ─┘

How to Dispatch

Use nats_request from your knight-comms skill for each sub-task:

// Parallel: fire all at once conceptually — but nats_request is blocking,
// so dispatch sequentially and collect. The knights process in parallel
// because each has its own consumer.

result_galahad = nats_request(
  knight: "galahad", domain: "security",
  task: "Summarize today's threat landscape. Focus on critical CVEs and active campaigns. Do not delegate to other knights.",
  timeout_ms: 300000
)

result_tristan = nats_request(
  knight: "tristan", domain: "infra",
  task: "Report cluster health: pod status, recent deployments, any alerts. Do not delegate to other knights.",
  timeout_ms: 180000
)

For sequential chains, include prior results in the next task:

result_galahad = nats_request(knight: "galahad", domain: "security",
  task: "Analyze CVE-2024-XXXX: affected software, severity, exploit status.")

result_tristan = nats_request(knight: "tristan", domain: "infra",
  task: "Based on this CVE analysis, check if we're affected:\n\n" + result_galahad)

Aggregation

After collecting results, synthesize them:

1. Extract key findings from each knight's response
2. Cross-reference — does Galahad's CVE affect what Tristan reported?
3. Identify action items — what needs to happen next?
4. Prioritize — what's urgent vs informational?
5. Format the unified report

Result Format

Structure your aggregated output as:

## ⚔️ Mission Report: [Title]

### 🎯 Action Items
- [ ] Critical items first
- [ ] Then important items

### 📊 Findings

#### [Knight] — [Domain]
Key findings from this knight...

#### [Knight] — [Domain]
Key findings from this knight...

### 🔗 Cross-References
- Finding X from Galahad relates to Finding Y from Tristan
- etc.

### 📝 Summary
One paragraph executive summary.

Knight Directory

Knight	Domain	NATS Subject	Specialty	Timeout
Galahad	security	fleet-a.tasks.security.>	Threat intel, CVEs, OpenCTI	300s
Kay	research	fleet-a.tasks.research.>	Deep research, news, media	300s
Tristan	infra	fleet-a.tasks.infra.>	Cluster health, Flux, k8s	180s
Lancelot	career	fleet-a.tasks.career.>	Interviews, professional	240s
Percival	finance	fleet-a.tasks.finance.>	Tax, budgets, Paperless	180s
Bedivere	home	fleet-a.tasks.home.>	Home Assistant, family	180s
Patsy	vault	fleet-a.tasks.vault.>	Obsidian vault curation	180s
Agravain	pentest	fleet-a.tasks.pentest.>	Offensive security, recon	300s
Gareth	wellness	fleet-a.tasks.wellness.>	Baby/family wellness	180s

Constraints

• Always include "Do not delegate to other knights" in sub-tasks to prevent infinite chains
• Cost awareness — each knight call costs ~$0.01-0.10. A 5-knight chain is ~$0.05-0.50
• Timeout — respect each knight's timeout. Don't send 300s tasks to 180s knights
• Concurrency — nats_request is blocking, but knights process independently. Order of dispatch matters for sequential chains, not for parallel ones.
• Error handling — if a knight times out, note it and continue with available results