Skip to content
Skills Library
AI & agents · Learning & explanation

agentic

/agentic

AI/LLM engineering patterns. Triggers when building agents, prompts, MCP servers, evals, or RAG. Model selection, caching, tool design, multi-agent orchestration.

Repository
codeforge ↗
Licence
MIT
Last repository change
26.06.2026

Assessment and security

95/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content30 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
2
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 26.06.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

AI/LLM engineering patterns. Triggers when building agents, prompts, MCP servers, evals, or RAG. Model selection, caching, tool design, multi-agent orchestration.

Original author description

AI/LLM engineering patterns. Triggers when building agents, prompts, MCP servers, evals, or RAG. Model selection, caching, tool design, multi-agent orchestration.

How it works

  1. Model selection by capability
  2. Prompt template (4 required fields)
  3. MCP tool rules
  4. Cost optimization
  5. Multi-agent pattern
Original SKILL.md headings · 59 lines, 329 words
SKILL.md preview
---
name: agentic
description: AI/LLM engineering patterns. Triggers when building agents, prompts, MCP servers, evals, or RAG. Model selection, caching, tool design, multi-agent orchestration.
allowed-tools: Read, Grep, Glob
---

# codeforge:agentic

## Model selection by capability

Don't pick by name. Pick by what the task needs. Names change; capabilities don't.

| Task | Capability needed | Why |
|------|-----------------|-----|
| Complex reasoning, code gen | MAX tier | Needs deep multi-step reasoning |
| High-volume classification, triage | FAST tier | Throughput over depth |
| Structured extraction | Any + tool use | Schema enforced by tool, not model |
| Long document analysis | STANDARD+ with large context | Context window is the constraint |
| Real-time / streaming response | FAST tier | Latency is the constraint |
| Adversarial verification | MAX tier | Needs strong self-critique |

For local models: start at STANDARD. Drop to FAST if context window < 32k or inference is slow.

## Prompt template (4 required fields)

```
## Role         specific grounded persona with domain expertise
## Task         exact outcome required, not vague instructions
## Constraints  hard limits, what NOT to do
## Output Format schema or example, never leave format ambiguous
```

Anti-patterns: "be helpful", "do your best", giant unstructured system prompts.

## MCP tool rules

| Rule | Detail |
|------|--------|
| Idempotent | Safe to retry without side effects |
| Schema first | description + JSON Schema on every tool |
| Structured errors | {error, code}, not raw stack traces |
| Minimal scope | Only permissions the tool actually needs |

## Cost optimization

| Technique | Saving |
|-----------|--------|
| Prompt caching on stable system prompts | ~90% on repeated calls |
| Batch API for non-real-time work | ~50% |
| Capability routing: FAST for triage, MAX for depth | 60-80% |
| Context compression on long sessions | Reduces input tokens |

## Multi-agent pattern

```
Orchestrator
├── Researcher  (read-only, Explore agent)
├── Implementer (write, isolated worktree)
└── Reviewer    (read-only, codeforge-reviewer)
```

Use `context: fork` for sub-agents that must not affect main context.
Use `isolation: worktree` for parallel file-writing agents.
Files in this package (2)
  • SKILL.md 2 KB
  • agents/openai.yaml 1 KB

Versions

The database has no other skills named agentic.

Database snapshot: 13.11.2026 · package version 0d516b216eca (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100ai
    Engineering OS playbook for building AI products and automation — LLM features, agents, tool use, RAG, MCP servers, structured outputs, prompt design, model selection, cost/latency, evals, and agentic automations (Claude API, Agent SDK, Vercel AI SDK, OpenAI, LangChain, Pydantic…
  • 90production-ai-agents
    Use when building an LLM agent, chatbot, RAG or tool-calling feature, choosing a model or provider, writing or changing prompts, managing context, latency or token cost, building LLM evals, or letting an AI change data.
  • 90ai-llm-patterns
    AI/LLM integration patterns: Anthropic SDK streaming with TypeScript, RAG architecture (chunking, embedding, vector search with pgvector), tool use / function calling, model selection guide (Haiku 4.5 vs Sonnet 5 vs Opus 4.8), prompt caching, structured output with Zod, token…
  • 100building-ai-agents
    Design, build, review or upgrade AI agents and LLM applications on a verified, current reference stack — model and SDK choice, context engineering, retrieval (RAG), MCP tools and A2A, memory, user-delegated authorization, prompt-injection security, agent UX, cost, observability,…
  • 90agentic-architecture
    Architecture of LLM agent systems: orchestration topologies (single agent, supervisor/sub-agents, pipelines, networks), memory/context strategy, the tool layer, and human-in-the-loop/control. Architect-level system design, not prompt wording. USE WHEN: designing…
  • 100llm-app-engineering
    Build reliable LLM-powered features: prompts, structured output, RAG, tool use, evals, cost and latency, safety. Use when integrating or reviewing LLM, agent or retrieval features in an application.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.