# LLM Runtime Architecture

> Use when designing how one canonical agent core runs across Codex, Claude Code, Gemini CLI, Cursor, and AGENTS.md-compatible tools.

- Skill: `agentlas-ai/llm-runtime-architecture` (Agent Skill)
- Install (CLI): `npx skillmds@latest add agentlas-ai/llm-runtime-architecture`
- Raw SKILL.md: https://api.skillmd.com/api/skills/agentlas-ai/llm-runtime-architecture/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: agentlas-ai (https://skillmd.com/u/agentlas-ai)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/agentlas-ai/llm-runtime-architecture

---


# LLM Runtime Architecture

## Procedure

1. Keep `AGENTS.md` as the canonical behavior contract.
2. For each runtime, name entry point, global command, adapter files, available
   tools, memory access, limitations, and verification command.
3. Keep adapters thin and point them back to the canonical core.
4. Write or repair `.agentlas/global-commands.json` when creating or packaging
   an agent.
5. State unsupported capabilities explicitly.

## Output

Return a runtime matrix with `runtime`, `entry_point`, `global_command`,
`adapter_files`, `memory_access`, `limitations`, and `verification`.