Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsAgentic Engineering Patterns Library Skill
Architectural library for building reliable, observable, and secure AI agents using proven engineering patterns.
$23.99
Agentic Engineering Patterns Library Skill
Example session with this skill installed
I'm building a customer support agent that uses a search tool and a refund tool. It keeps looping when it can't find an order. Help me harden the architecture.
- Read your context and instructions
- Compiled the agentic engineering patterns
I have analyzed your support agent's failure mode. The infinite loop stems from a lack of exit conditions in the reasoning cycle and unhandled tool errors. I recommend implementing a Bounded Agent Loop paired with a Circuit Breaker for the refund tool.
Agent Hardening Architecture
| Pattern | Failure Mode Addressed | Implementation Logic |
|---|---|---|
| Bounded Agent Loop | Infinite reasoning loops | Set max_iterations=5 and max_time_ms=30000. Exit to human handoff on limit reach. |
| Circuit Breaker | Repeated refund tool failures | Track 5xx errors; if threshold > 3, disable tool for 5 minutes and return fallback message. |
| Preflight Checks | Invalid refund attempts | Validate order ID format and status (e.g., 'shipped') via a typed schema before calling tool. |
| Human Handoff Gate | Ambiguous order search | If search returns 0 results twice, escalate to a human agent with the current conversation trace. |
Implementation Recipe: Bounded Loop
# Pseudocode for hard-coded safety boundary
def execute_agent(task):
state = {"iteration": 0, "max": 5, "status": "active"}
while state["iteration"] < state["max"]:
result = model.run(task, tools=[search, refund])
if result.complete:
return verify(result)
state["iteration"] += 1
return escalate_to_human("Iteration limit reached without resolution.")
Next steps
- Define a strict JSON schema for the refund tool to prevent the model from hallucinating order IDs.
- Add a
Postcondition Checkto verify the refund status in the database after the tool reports success. - Create a small 'Golden Set' of failing order IDs to test the new escalation logic.
Connects securely to your tools. The creator never sees your data.
About this skill
Building AI agents that actually work in production requires moving beyond simple prompts to hardened engineering patterns. This skill provides a library of battle-tested architectural patterns designed to solve common agent failure modes like infinite loops, tool hallucination, and context leakage. It helps you design systems that are observable, secure, and maintainable by applying structured control flows and rigorous validation logic.
What it does
- Architects control flow using bounded loops, state machines, and supervisor-worker patterns to prevent runaway execution.
- Hardens tool integration with typed contracts, least-privilege scoping, and circuit breakers for external API calls.
- Optimizes context management through budget allocation, evidence-carrying retrieval, and untrusted content isolation.
- Implements reliability patterns such as structured output validation, independent verification, and golden set regression testing.
- Enforces security boundaries via approval gates for side effects, PII minimization, and tenant isolation.
How it works
- Define the system constraints including failure impact, model choice, available tools, and success criteria.
- Classify failure modes across quality, control-flow, safety, or operational categories to identify technical gaps.
- Select architectural patterns from the library that map specifically to your identified risks.
- Generate implementation cards providing pseudocode, sequence diagrams, failure mitigations, and test cases for each pattern.
Frameworks & tools
This skill is platform-agnostic and applies to LangChain, LlamaIndex, CrewAI, AutoGPT, or custom TypeScript/Python agent stacks. It focuses on the architectural logic rather than specific library syntax.
Why this beats prompting it yourself
Prompting an agent to "be reliable" fails because reliability is a system property, not a linguistic instruction. This skill provides the structural scaffolding—like state machines and validation gates—that natural language alone cannot enforce.
Use cases
- Financial agents requiring explicit human-in-the-loop gates for high-value transactions.
- Multi-agent swarms needing structured handoff contracts and shared-state ownership.
- Data extraction pipelines requiring claim-evidence mapping and source provenance tags.
- Customer support bots needing circuit breakers to stop repetitive failing tool calls.
Known limitations
Does not provide production-ready code for specific proprietary SDKs. Requires engineering judgment to determine which patterns fit your specific latency and cost budgets.
Architectural guidance only; does not provide library-specific boilerplate. Logic requires manual implementation. Effectiveness depends on model capabilities and runtime environment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
Permissions
No special permissions declared or detected