Agent Plan Writing

Agent Plan Writing

Overview

Agent plan writing is the discipline of designing execution plans for AI agent workflows. The harness matters more than the model. Agent completion rates depend more on action-space design, context engineering, and orchestration patterns.

Output format

When this skill activates, produce a markdown agent plan containing:

1. Workflow Overview — what the system does, which orchestration pattern, and why
2. Agent Roster — each agent’s role, model, tools, and context scope
3. Orchestration Graph — how agents coordinate
4. Context Budget — token allocation per agent
5. Safety Constraints — permission boundaries, output validation, human-in-the-loop gates
6. Evaluation Plan — what to trace, quality metrics
7. Failure Handling — per-pattern failure modes and recovery strategies

Orchestration patterns

Five patterns dominate production agent systems:

• Fan-Out: Parallel execution of independent subtasks. Coordinator dispatches to N agents simultaneously.
• Pipeline: Sequential chain where each stage requires the prior stage’s output.
• Supervisor: A supervisor agent decomposes the task, delegates to specialists, and synthesizes results. The 2026 production default.
• Debate: Multiple agents reason independently, then argue toward convergence.
• Swarm: Dynamic spawning of agents based on workload.

Context window budget planning

Agents consume ~7x more tokens than standard chat sessions. Plan token budgets explicitly.

Budget allocation:

1. System prompt: 500–2,000 tokens. Cached input costs 10–25% of normal.
2. Tool schemas: Each MCP tool adds 100–500 tokens to context.
3. Working memory: Reserve 30–50% of context for conversation/reasoning accumulation.
4. Output headroom: Reserve 15–25% for the agent’s response generation.

Safety guardrails in agent plans