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/szara7678/openakashic/agent-design-patternsgit clone --depth 1 https://github.com/szara7678/OpenAkashicWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/szara7678/openakashic/agent-design-patterns)<a href="https://agentmods.dev/agents/szara7678/openakashic/agent-design-patterns"><img src="https://agentmods.dev/badge/agents/szara7678/openakashic/agent-design-patterns.svg" alt="Measured on agentmods" height="20"></a>What 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 5.1 | $0.00000 | $0.02974 |
| Opus 5 | $0.00000 | $0.01487 |
| Sonnet 5 | $0.00000 | $0.00595 |
| Haiku 4.5 | $0.00000 | $0.00297 |
Grade A, and why
agent-design-patterns 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 5d ago.
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 — 295 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Summary
LLM 에이전트 설계 패턴 레퍼런스. Anthropic "Building Effective Agents" 5가지 패턴 + 실전 적용 기준. 워크플로우 vs 에이전트 구분, 패턴별 트레이드오프, 프로덕션 배포 체크리스트. 2025 기준.
Sources
- Anthropic "Building Effective Agents" (2024)
- Anthropic Claude Agent SDK 문서
- AI Engineer Summit 2024 발표 내용
- LLM Agent 실전 운영 경험 종합
1. 핵심 개념: 워크플로우 vs 에이전트
워크플로우 (Workflow)
- LLM 호출 경로가 코드로 미리 정해져 있음
- 예: A → B → C 순서 고정
- 예측 가능, 디버그 쉬움
- 단순·반복·고정 로직에 적합
에이전트 (Agent)
- LLM이 스스로 다음 행동을 결정
- 도구를 언제·어떻게 쓸지 모델이 판단
- 유연하지만 비결정론적
- 복잡·모호·탐색이 필요한 태스크에 적합
선택 기준: "LLM 없이도 같은 로직을 if/else로 표현할 수 있다면 워크플로우."
2. 5가지 설계 패턴
2-1. 프롬프트 체이닝 (Prompt Chaining)
Input → [LLM 1] → 중간 출력 → [LLM 2] → 최종 출력
언제: 태스크가 명확히 분리된 순차적 단계로 구성될 때.
- 글 초안 작성 → 문체 교정 → 번역
- 코드 생성 → 테스트 작성 → 문서화
장점: 단계별 검증 가능. 각 LLM 호출의 컨텍스트를 좁힐 수 있음. 단점: 오류가 하위 단계로 전파됨. 전체 지연 시간 = 각 단계 합산.
def chain(input_text):
draft = llm("초안 작성: " + input_text)
edited = llm("문체 교정: " + draft)
translated = llm("한국어 번역: " + edited)
return translated
2-2. 라우팅 (Routing)
Input → [분류 LLM] → 경로 A / 경로 B / 경로 C
언제: 입력 타입에 따라 전혀 다른 처리가 필요할 때.
- 고객 문의 → 기술 지원 / 결제 문의 / 일반 문의
- 코드 → 언어별 특화 모델 (Python vs Go vs Rust)
장점: 각 경로를 독립적으로 최적화. 복잡한 시스템을 전문화된 서브시스템으로 분해. 단점: 분류 오류 시 전체 실패. 경계가 모호한 케이스 처리 필요.
def route(input_text):
category = llm(f"분류 (tech/billing/general): {input_text}")
handlers = {
"tech": handle_tech,
"billing": handle_billing,
"general": handle_general,
}
return handlers.get(category, handle_general)(input_text)
2-3. 병렬화 (Parallelization)
두 가지 하위 유형:
섹셔닝 (Sectioning): 독립적인 서브태스크를 동시 실행
Input → [LLM A] ─┐
→ [LLM B] ─┤→ 집계 → Output
→ [LLM C] ─┘
투표 (Voting): 동일 태스크를 여러 번 실행 후 다수결
Input → [LLM 1] ─┐
→ [LLM 2] ─┤→ Majority Vote → Output
→ [LLM 3] ─┘
언제 섹셔닝: 긴 문서를 청크별로 분석, 여러 관점(보안/성능/유지보수) 동시 평가. 언제 투표: 정확도가 중요한 분류, 코드 보안 검토, 의료 진단 보조.
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.
- 5d ago First seen · 295 lines · 0 tokens per session scan A 8f4495fa6127
agent-design-patterns is an agent published in the GitHub repository szara7678/OpenAkashic (3 stars, last pushed 2mo ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 2,974 tokens. 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
gsd-planner
Creates executable phase plans with task breakdown, dependency analysis, and goal-backward verification. Spawned by $gsd-plan-phase orchestrator.
brain-os-mode
Use for any Brain OS task delegated to a subagent — reading entity state, checking active decisions, running focus/patterns/retro analysis, or proposing changes that need to honor active decisions. Always uses Brain OS MCP tools (mcpbrain-os) first; falls back to file reads only when the MCP server is unreachable.…
release_agent
Agent with permissions to release Agent Brain packages. Handles version bumping, quality gates, building wheels, git tagging, GitHub release creation, and PyPI publish verification. Use when running /ag-brain-release or any release workflow.
socratic_mentor_agent
Guides paper authors through Socratic questions to sharpen arguments and surface unstated assumptions.
document-generator
Generate documents, reports, and slides. Learns style patterns and template structures.
uninstall_agent
Agent with permissions to remove Agent Brain local installations.