--- [EXPLICIT] name: prompt-creator description: [EXPLICIT] Generates structured prompts for 9 prompt types in agentic ecosystems — meta, handoff, deliberation, [EXPLICIT] synthesis, validation, and fallback. Activates when the user says "create a prompt", "write a handoff prompt", [EXPLICIT] "generate a meta prompt", "build a committee deliberation prompt", or "make a fallback prompt". Also triggers [EXPLICIT] on mentions of prompt types, agent prompts, or multi-agent prompt design. Use this skill even if the user [EXPLICIT] only names the agent — it interviews for prompt type and context. [EXPLICIT] argument-hint: prompt-type owning-agent-id [EXPLICIT] model: opus [EXPLICIT] context: fork [EXPLICIT] allowed-tools: Read, Write, Edit, Glob, Grep [EXPLICIT]

Prompt Creator

Generate structured prompt files for multi-agent ecosystems. Covers 9 prompt types from system prompts to committee deliberation to fallback recovery. [EXPLICIT]

Assumptions & Limits

• Assumes an agentic ecosystem with defined agents (prompts reference agent constitutions)
• Limit: Prompts are templates, not runtime — placeholders ({{var}}) are filled by the orchestrator
• Limit: Committee prompts (deliberation/synthesis) require ≥3 agents to be meaningful
• Trade-off: More detailed prompts = more predictable agent behavior but less adaptability. For creative tasks, keep prompts loose.

Usage

/prompt-creator meta_prompt data-analyst
/prompt-creator handoff_prompt customer-onboarding
/prompt-creator committee_deliberation                # interview mode

Parse $1 as prompt type, $2 as owning agent ID. If missing, ask. [EXPLICIT]

Before Generating

1. Read the agent: Read agents/$2/agent.md — ground prompt in agent identity, tone, and constraints
2. Check existing prompts: Glob agents/$2/prompts/*.md — avoid duplicates
3. Read prompt spec: Read references/prompt-types-spec.md if available

The 9 Types

#	Type	File Pattern	Purpose	Complexity
1	agent_system_prompt	agent.md	Full constitution	→ Use /agent-constitution-creator
2	meta_prompt	prompts/meta-{topic}.md	Behavioral instruction for ONE aspect	Low
3	system_user_pair	prompts/pair-{scenario}.md	Reusable system+user template	Low
4	workflow_step_prompt	Inline in skill.yaml	Step-level LLM instruction	→ Use /workflow-creator
5	handoff_prompt	prompts/handoff.md	Task transfer protocol	Medium
6	committee_deliberation	prompts/deliberation.md	Independent multi-agent evaluation	High
7	committee_synthesis	prompts/synthesis.md	Merge multiple agent responses	High
8	validation_prompt	prompts/validation.md	Quality validation of outputs	Medium
9	fallback_prompt	prompts/fallback.md	Recovery when primary fails	Medium

Types 1 and 4 redirect to specialized skills — this skill handles types 2, 3, 5-9. [EXPLICIT]

Output Format

Write to agents/{agentId}/prompts/{filename}.md: [EXPLICIT]

---
type: "{promptType}"
owningAgent: "{agentId}"
sourceAgentMd: "agents/{agentId}/agent.md"
version: "1.0.0"
---

# {Title}

{Content per type-specific rules below}

Type-Specific Rules

meta_prompt — Behavioral aspect instruction

• Focus: Exactly ONE aspect: reasoning OR formatting OR restrictions OR style
• Structure: Preamble (agent identity) → Framework (the rules) → Constraints (boundaries)
• Anti-pattern: Combining reasoning + formatting in one meta_prompt → split into two

# Reasoning Meta-Prompt

You are {{agent.name}}. Your role: {{agent.role}}. [EXPLICIT]

## Framework
1. {Step 1 of reasoning process}
2. {Step 2}

## Constraints
- {Hard limit 1}
- {Hard limit 2}

system_user_pair — Reusable scenario template

• Must have: ## System and ## User sections
• System: Sets context, constraints, output format
• User: Scenario template with {{placeholders}}
• Design rule: Each pair handles ONE scenario (not a Swiss Army knife)

handoff_prompt — Task transfer protocol

• Must specify:
  • Context to PASS: task state, progress, relevant data
  • Context to OMIT: internal reasoning, failed attempts, irrelevant history
  • Target agent: explicit ID
  • Success criteria: how target agent knows it's done
• Anti-pattern: Passing the entire conversation (context explosion)

committee_deliberation — Independent evaluation

• Must require: Agent gives INDEPENDENT opinion FIRST, before seeing others
• Must include: Scoring rubric with weighted dimensions
• Must specify: Output format for structured comparison
• Key insight: The value of committees comes from independent evaluation — if agents see each other's work first, they converge prematurely

committee_synthesis — Multi-response merger

• Must define: Redundancy removal strategy, conflict resolution method, confidence weighting
• Merge strategies:
  • Majority vote (for binary decisions)
  • Weighted average (for numeric assessments)
  • Reasoned selection (for qualitative choices — requires justification)

validation_prompt — Quality checker

• Must define: Pass/fail criteria with severity levels (critical/major/minor)
• Must produce: Actionable feedback (not "this could be better" but "section 3 missing required field X")
• Must reference: The DoD/qaChecklist from the originating workflow

fallback_prompt — Recovery playbook

• Must define: Trigger conditions (when does fallback activate?)
• Must specify: Preservation priorities (what to save vs sacrifice)
• Must include: User communication template (how to explain the degradation)
• Must have: Escalation path if fallback also fails

Example: Good vs Bad

Bad handoff_prompt:

Hand off the task to the next agent. [EXPLICIT]

Good handoff_prompt:

## Handoff Protocol: {{source_agent}} → {{target_agent}}

### Context to Pass
- Task ID: {{task_id}}
- Current state: {{state_summary}}
- Completed steps: {{completed_steps}}
- Pending decision: {{decision_needed}}

### Context to Omit
- Internal reasoning chains
- Failed approaches and why they failed
- Intermediate calculations

### Success Criteria
The handoff is complete when {{target_agent}} confirms: [EXPLICIT]
- [ ] Context received and understood
- [ ] Can proceed without clarification
- [ ] Estimated completion: {{time_estimate}}

Validation Gate

• YAML frontmatter has type, owningAgent, sourceAgentMd, version
• Type is one of the 9 defined types (or redirects to appropriate skill)
• Agent.md was read and prompt references agent identity
• Type-specific rules followed (see table above)
• No empty sections
• All {{placeholders}} are named descriptively (not {{x}})
• Committee prompts require independent evaluation before comparison
• Handoff prompts specify both what to pass AND what to omit
• Validation prompts include severity levels
• Fallback prompts include escalation path

---

Author: Javier Montaño | Last updated: March 12, 2026

Edge Cases

Scenario	Handling
Empty or minimal input	Request clarification before proceeding
Conflicting requirements	Flag conflicts explicitly, propose resolution
Out-of-scope request	Redirect to appropriate skill or escalate