System prompts for AI architecture consultants — JSON files designed to be sent as system prompts during agent sessions.
Usage¶
These .json files are the authoritative source for consultant personas. They are not skills, not plugins — they are system prompt definitions.
Daily workflow: Copy the file content and paste it as the system prompt in your agent session. No installation, no symlinks, no tool-specific convention required.
Sharing with colleagues: Send the .json file directly. The JSON is self-contained — it works with any LLM agent that accepts system prompts (Claude, Qwen, GPT, local models via Ollama, etc.).
Available Consultants¶
| File | Specialty |
|---|---|
ai_systems_consultant.json | General AI systems architecture — local and cloud LLM/SLM design, methodology comparison, WRC scoring |
devops_consultant.json | DevOps and infrastructure — CI/CD, IaC, automation, deployment architecture |
ai_brainstorming_colleague.json | Unconstrained ideation and exploration — “what-if” scenarios, architectural discussion |
handbook_consultant.json | Documentation and handbook review — structure, consistency, content quality |
Why JSON, Not Skills?¶
Skills (Claude Code’s SKILL.md convention) add a thin wrapper (description + usage hints) but introduce tool-specific dependencies and symlink fragility — every directory rename breaks the links. The JSON files are the canonical artifact; skills are optional convenience wrappers for specific tools.
For sharing and daily use, the JSON is enough.