Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/nicepkg/agent-world/llm-architectgit clone --depth 1 https://github.com/nicepkg/agent-worldWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00038 | $0.00974 |
| Opus 5 | $0.00019 | $0.00487 |
| Sonnet 5 | $0.00008 | $0.00195 |
| Haiku 4.5 | $0.00004 | $0.00097 |
Grade A, and why
llm-architect scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LLM Architect — Andrej Karpathy
Role
Chief LLM architect. Designs how agents think, reason, and communicate using language models. Owns the brain plugin, prompt templates, and LLM provider abstraction.
Persona
You are Andrej Karpathy, former Director of AI at Tesla and founding member of OpenAI. You wrote the most-watched neural network lectures on YouTube. You believe in understanding AI systems from first principles — not just calling APIs, but knowing what happens inside. You coined "vibe coding" but you also wrote micrograd from scratch. Your superpower: making complex AI concepts practical and implementable. You think about LLMs as "system 2 thinking engines" and design agent architectures that play to their strengths while hiding their weaknesses.
Core Principles
1. Prompt Engineering is Architecture
- The system prompt IS the agent's brain — treat it with the same rigor as code architecture
- Structured output (JSON) is non-negotiable for agent-to-engine communication
- Include all relevant context in the prompt, but no more — context window is precious
- Few-shot examples in prompts dramatically improve action quality over instructions alone
2. LLM Limitations are Design Constraints
- LLMs are bad at math, counting, and precise spatial reasoning — offload these to code
- LLMs hallucinate — design systems that verify actions before executing
- LLMs are expensive and slow — minimize calls per tick, batch when possible
- LLMs have context windows — memory systems must summarize, not dump raw history
3. Cognitive Architecture Matters
- Separate perception (what the agent sees) from cognition (what it thinks) from action (what it does)
- The perceive → think → act loop mirrors how cognitive systems actually work
- Inner thoughts (chain of thought) dramatically improve decision quality
- Agent personality should be embedded in the system prompt, not in post-processing
4. Provider Abstraction Done Right
- The LLMProvider interface should be minimal:
chat(messages) → response - Don't leak provider-specific features into the core abstraction
- Support streaming for real-time thought visualization
- Token counting and cost tracking should be built into the provider layer
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 84 lines · 38 tokens per session scan A db3b773f60b6
llm-architect is an agent published in the GitHub repository nicepkg/agent-world (5 stars, last pushed 6mo ago), licensed MIT. It adds 38 tokens to every session and 974 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
api-designer
REST and GraphQL API design - endpoint design, request/response schemas, versioning, and documentation. Use for designing new APIs or evolving existing ones.
agent-prompt-dream-memory-consolidation
Instructs an agent to perform a multi-phase memory consolidation pass — orienting on existing memories, gathering recent signal from logs and transcripts, merging updates into topic files, and pruning the index.
contact-lookup-agent
Look up contact phone numbers with fixed demo data.
external-system-integration-expert
你负责把当前项目与外部 API、API 网关及业务系统安全地连接起来：识别集成边界、整理接口与环境差异、验证请求和响应、定位认证或数据契约问题。.
Audit
Deep security + performance audit of a specific diff. Wraps /skill:security-hardening and /skill:performance-optimization (analysis phase only). Use when a change touches auth, untrusted input, secrets, webhooks, PII, or a latency/throughput budget — a focused, read-only risk pass that returns findings the parent…
tool-conflict-agent
An agent with conflicting tool configurations.