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 skills add hoangatg/ai-agent-toolkit --skill llm-app-patternsgit clone --depth 1 https://github.com/hoangatg/ai-agent-toolkitWrote 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/skills/hoangatg/ai-agent-toolkit/llm-app-patterns)<a href="https://agentmods.dev/skills/hoangatg/ai-agent-toolkit/llm-app-patterns"><img src="https://agentmods.dev/badge/skills/hoangatg/ai-agent-toolkit/llm-app-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 | $0.00040 | $0.01120 |
| Opus 5 | $0.00020 | $0.00560 |
| Sonnet 5 | $0.00008 | $0.00224 |
| Haiku 4.5 | $0.00004 | $0.00112 |
Grade A, and why
llm-app-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 4d 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 — 179 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LLM App Patterns
Ship LLM features that are fast, reliable, and cost-effective in production.
1. Architecture Decisions
Integration Patterns
| Pattern | Use Case | Complexity |
|---|---|---|
| Direct API | Simple chat, single model | Low |
| Gateway/Proxy | Multi-model, rate limiting | Medium |
| Queue-based | High throughput, async tasks | Medium |
| Agent framework | Complex reasoning, tool use | High |
Model Selection
| Factor | Consideration |
|---|---|
| Task complexity | Simple → small model, complex → large |
| Latency | Streaming for UX, batch for backend |
| Cost | Per-token pricing, caching potential |
| Privacy | Cloud vs self-hosted |
| Reliability | Uptime SLA, fallback models |
2. Streaming Patterns
When to Stream
| Scenario | Stream? |
|---|---|
| User-facing chat | ✅ Always |
| Background processing | ❌ Batch |
| Function calling | ⚠️ Depends on framework |
| Structured output | ❌ Wait for complete JSON |
Implementation Principles
- Use Server-Sent Events (SSE) for web
- Buffer partial tokens for smooth display
- Handle stream interruptions gracefully
- Implement cancel/abort mechanisms
3. Caching Strategies
Cache Layers
| Layer | What to Cache | TTL |
|---|---|---|
| Prompt cache | Full prompt → response | Hours-Days |
| Semantic cache | Similar queries → cached response | Hours |
| Embedding cache | Text → vector | Days-Weeks |
| Response cache | API response → result | Minutes-Hours |
Cache Decision
Should you cache?
├── Deterministic output? → Exact match cache
├── Similar queries common? → Semantic cache
├── Expensive computation? → Result cache
└── High latency API? → Response cache
4. Error Handling & Resilience
Failure Modes
| Failure | Strategy |
|---|---|
| Rate limited | Exponential backoff + queue |
| Timeout | Set deadline, return partial |
| Model down | Fallback to alternative model |
| Bad output | Retry with rephrased prompt |
| Token overflow | Truncate context intelligently |
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.
- 4d ago First seen · 179 lines · 40 tokens per session scan A 89035cb8adb4
llm-app-patterns is a skill published in the GitHub repository hoangatg/ai-agent-toolkit (1 stars, last pushed 5mo ago), licensed MIT. It adds 40 tokens to every session and 1,120 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 2026-09-03.
Other skills, from other repositories
opencode
Delegate coding tasks to the OpenCode CLI for feature work, refactoring, PR review, and autonomous-style runs. Requires opencode installed where the agent can execute shell commands.
codex
Delegate coding tasks to the OpenAI Codex CLI for features, refactoring, PR reviews, and batch fixes. Requires the codex CLI and typically a git repository.
data-pipelines
Apply Data Pipelines Pocket Reference practices (James Densmore). Covers Infrastructure (Ch 1-2: warehouses, lakes, cloud), Patterns (Ch 3: ETL, ELT, CDC), DB Ingestion (Ch 4: MySQL, PostgreSQL, MongoDB, full/incremental), File Ingestion (Ch 5: CSV, JSON, cloud storage), API Ingestion (Ch 6: REST, pagination, rate…
deepnote
DeepNote knowledge base: persistent interlinked markdown wiki with ingest, query, lint, link graph and history.
golang-patterns
Idiomatic Go patterns, best practices, and conventions for building robust, efficient, and maintainable Go applications.
clickhouse-io
ClickHouse database patterns, query optimization, analytics, and data engineering best practices for high-performance analytical workloads.