get_system_prompt

Generate a dynamic system prompt for an MCP consumer based on the current engagement state.

Purpose

Returns a markdown system prompt tailored to the specified role, including engagement briefing, tool reference table, state snapshot, and OPSEC constraints. Use this instead of static AGENTS.md instructions for session initialization and agent dispatch.

Parameters

Parameter	Type	Required	Default	Description
role	'primary' \| 'sub_agent'	Yes	—	Consumer role: primary orchestrator or scoped sub-agent
agent_id	string	No	—	For sub_agent role: the agent ID to scope the instructions
include_state	boolean	No	true	Include current state snapshot in the prompt
include_tools	boolean	No	true	Include tool reference table in the prompt

Roles

primary

Full orchestrator instructions including: - Engagement briefing (name, scope, objectives, OPSEC profile) - Core loop workflow (get_state → next_task → validate → execute → report) - Key principles - Complete tool reference table - Current state snapshot (graph summary, access level, frontier size)

sub_agent

Scoped worker instructions including: - Engagement name and agent context - Scoped tool subset (get_agent_context, validate_action, report_finding, etc.) - Step-by-step workflow - Agent-specific context (frontier item, scoped nodes, skill)

Example

{
  "role": "primary",
  "include_state": true,
  "include_tools": true
}

Returns

A single text content block containing the generated markdown system prompt.

Annotations

• Read-only: Yes
• Idempotent: Yes