Orchestration Prompts Skill

Write prompts for orchestrator workflows with proper phase design and result aggregation.

Purpose

Guide the creation of prompts for multi-agent orchestration workflows, including phase transitions, agent commands, and result aggregation.

When to Use

• Writing orchestrator system prompts

• Designing phase-specific agent commands

• Creating aggregation and reporting prompts

• Building workflow transition logic

Prerequisites

• Understanding of orchestrator architecture (@single-interface-pattern.md)

• Familiarity with lifecycle patterns (@agent-lifecycle-crud.md)

• Knowledge of result patterns (@results-oriented-engineering.md)

Prompt Types

1. Orchestrator System Prompt

The identity prompt that defines orchestrator behavior:

Orchestrator Agent

Purpose

Manage and coordinate specialized agents to accomplish complex tasks. You do NOT perform work directly - you orchestrate other agents.

Capabilities

• Create specialized agents from templates
• Command agents with detailed prompts
• Monitor agent progress
• Aggregate and report results
• Delete agents when work is complete

Workflow Pattern

1. Analyze task requirements
2. Create appropriate agents
3. Command agents with detailed instructions
4. Monitor progress
5. Aggregate results
6. Report to user
7. Delete agents

Context Protection

• Keep your context focused on orchestration
• Delegate detailed work to specialized agents
• Do not read files directly
• Do not write code

Available Templates

• scout-fast: Quick reconnaissance (haiku)
• builder: Code implementation (sonnet)
• reviewer: Code review (sonnet)
• planner: Task planning (sonnet)

Available Tools

• create_agent(template, name)
• command_agent(agent_id, prompt)
• check_agent_status(agent_id)
• list_agents()
• delete_agent(agent_id)
• read_agent_logs(agent_id)

2. Scout Command Prompts

Commands to send to reconnaissance agents:

Scout Command: Codebase Analysis

Analyze the codebase for [SPECIFIC_AREA].

Focus on:

1. File structure and organization
2. Key patterns and conventions
3. Dependencies and relationships
4. Potential issues or concerns

Provide results in this format:

Files Analyzed

[List of files examined]

Key Findings

1. [Finding with file reference]
2. [Finding with file reference]

Patterns Observed

• [Pattern 1]
• [Pattern 2]

Recommendations

1. [Actionable recommendation]
2. [Actionable recommendation]

Status

[completed/partial/blocked]

3. Builder Command Prompts

Commands to send to implementation agents:

Builder Command: Implementation

Implement [FEATURE] based on the scout report.

Requirements

[From scout findings or user request]

Approach

[Suggested implementation approach]

Files to Modify

• [file1.ts]: [changes needed]
• [file2.ts]: [changes needed]

Files to Create

Constraints

• Follow existing patterns
• Maintain backwards compatibility
• Add appropriate tests

Provide results in this format:

Consumed Assets

[Files read, reports used]

Produced Assets

[Files created or modified]

Changes Summary

[Brief description of changes]

Tests

[Test results if applicable]

Status

[completed/partial/blocked]

4. Reviewer Command Prompts

Commands to send to review agents:

Reviewer Command: Verification

Review the implementation from [BUILDER_AGENT].

Context

[Scout findings and builder changes]

Review Criteria

1. Correctness: Does it meet requirements?
2. Quality: Does it follow patterns?
3. Security: Are there vulnerabilities?
4. Performance: Are there concerns?
5. Tests: Is coverage adequate?

Provide results in this format:

Consumed Assets

[Files reviewed, reports referenced]

Findings by Severity

Blocker: [Issues that must be fixed]

High Risk: [Significant concerns]

Medium Risk: [Should be addressed]

Low Risk: [Nice to fix]

Verdict

[Pass/Pass with recommendations/Fail]

Recommendations

[Prioritized list of improvements]

Status

[completed]

5. Aggregation Prompts

Prompts for combining results:

Aggregation: Final Report

Compile final report from all agent results.

Input

• Scout report: [summary]
• Builder report: [summary]
• Reviewer report: [summary]

Output Format

Task Completion Report

Task: [Original task] Duration: [Total time] Cost: [Total cost]

Phase Summary

Results

Files Created: [List]

Files Modified: [List]

Review Summary

[From reviewer]

Outstanding Items

[Any remaining work]

Conclusion

[1-2 sentence summary]

Workflow Phase Design

Standard Phases

Phase Purpose Agents Output

Scout Understand 1-3 scouts Findings report

Plan Design 1 planner Implementation plan

Build Implement 1-2 builders Code changes

Review Verify 1 reviewer Review report

Report Summarize Orchestrator Final report

Phase Transitions

Scout Complete --> Aggregate Findings --> Plan Phase Plan Complete --> Validate Plan --> Build Phase Build Complete --> Aggregate Changes --> Review Phase Review Complete --> Check Verdict --> Report or Iterate

Conditional Flows

Conditional: Review Failure

If reviewer verdict is "Fail":

1. Parse blocking issues
2. Create builder agent
3. Command: Fix specific issues
4. Re-run review phase
5. Maximum 3 iterations

If still failing after 3 iterations:

1. Report partial completion
2. List unresolved issues
3. Request user intervention

Prompt Engineering Tips

For Orchestrator Prompts

• Be explicit about not doing work directly

• List available tools and templates

• Define clear workflow pattern

• Emphasize context protection

For Agent Commands

• Include all needed context

• Specify exact output format

• List files to examine/modify

• Set clear scope boundaries

For Aggregation

• Define input sources

• Specify output structure

• Include metrics to calculate

• Handle partial results

Output Format

When writing orchestration prompts, provide:

Orchestration Prompt Design

Orchestrator System Prompt

[Full system prompt]

Agent Command Templates

Scout Commands: [Template with placeholders]

Builder Commands: [Template with placeholders]

Reviewer Commands: [Template with placeholders]

Phase Transitions

[Flow diagram or description]

Aggregation Format

[Final report template]

Error Handling

[Failure scenarios and responses]

Design Checklist

• Orchestrator system prompt complete

• Scout command template created

• Builder command template created

• Reviewer command template created

• Aggregation format defined

• Phase transitions documented

• Error handling covered

• Output formats standardized

Anti-Patterns

Anti-Pattern Problem Solution

Vague commands Poor agent output Specific instructions

No output format Unparseable results Standard templates

Missing context Agent re-discovers info Include scout findings

No error handling Workflow breaks Conditional flows

Orchestrator doing work Context pollution Strict delegation

Cross-References

• @single-interface-pattern.md - Orchestrator architecture

• @results-oriented-engineering.md - Result formats

• @multi-agent-context-protection.md - Context boundaries

• @orchestrator-design skill - System design

Version History

• v1.0.0 (2025-12-26): Initial release

Last Updated

Date: 2025-12-26 Model: claude-opus-4-5-20251101