trpg

TRPG Writing

Sub-skill of general-writing. Inherits shared principles, anti-AI discipline, and quality gates.

CRITICAL — Review → Fix Brief → Dispatched Subagent: When a review finds issues, translate them into a detailed fix brief (file paths, quoted findings, expected outcome, scope boundary, self-check) and dispatch a fresh subagent to apply the fix. Never use inline sed/edit_file fixes.

CRITICAL — Anti-Read-Loop Rule: The orchestrator must dispatch its first subagent (world_designer, encounter_designer, or npc_writer) within 10 iterations. If the orchestrator has spent 10+ iterations only reading files without creating any subagent, it is in a read-loop. Stop reading, write your plan from what you have, and dispatch immediately.

CRITICAL — Step Budget & Delivery: You have a limited step budget. Plan delivery accordingly:

• One-shots: fully achievable in one session. Write all components, review, assemble, convert to .docx.
• Campaigns / multi-session modules: prioritize core scenario + key NPCs + encounter maps. Deliver completed components with .docx and ask user whether to continue.
• Before your session ends: always assemble whatever is done and convert to .docx — never end with only raw component files.

Trigger Keywords

跑团, TRPG, tabletop RPG, CoC, DnD, 模组, scenario, one-shot, campaign, keeper, dungeon master, NPC设计, encounter, Pathfinder, FATE, sandbox, 战役, 调查员, 修仙

File Paths

Output Directory Structure

/mnt/agents/output/
├── research/                    # User uploads (system rulebooks, setting refs)
├── {project_name}.outline.md
├── documents/
│   ├── {project_name}.npcs.md
│   ├── {project_name}.encounters.md
│   └── {project_name}.maps.md
└── scenes/
    ├── {project_name}.scene{NN}.md
    ├── {project_name}.stat_blocks.md
    └── {project_name}.handouts/

User-Uploaded Files

Before Stage 1, copy all user-uploaded files (system rulebooks, setting references, existing campaign notes) to /mnt/agents/output/research/. Preserve original filenames. Reference these copies (not originals) in all subagent task prompts — subagents may run in sandboxed environments without access to the original paths.