cc-skill:agent-sdk-patterns

Claude Code skill · anthropic · active

last 30 days

0 calls

0 of 30 days active

explain like I am new to this

what it does

It gives you the patterns and template code to build autonomous agents that can use tools, think through multi-step problems, and execute actions. The skill handles tool definitions, retry logic, cost tracking, and safety limits so your agent doesn't spiral into costly loops or give up too early.

who it is for

Developers building workflows where an AI needs to make decisions, fetch data from databases, call APIs, or take multiple steps to solve a problem rather than just generate one output.

when it fires

When you mention building an agent, multi-step AI workflow, tool-using system, agent loop, autonomous agent, or agent pipeline.

say something like

"Build me an agent that reads our weekly metrics and writes a memo"
"I need an autonomous workflow that decides what content to post and generates it"
"Create an agent loop that can query the database and make decisions based on what it finds"

what you might get back

A TypeScript template for a Supabase Edge Function showing how to define tools (like read_supabase, insert_supabase), set up Claude with those tools, handle tool responses in a loop, track costs, and stop after 10 turns. It includes error handling and retry logic built in.

how to know it worked

You get a working TypeScript agent template with defined tools, a message loop that handles Claude's responses, and guardrails like turn limits and cost logging already in place.

is this skill working for me

never used

cloud runs

0 deployments

local fires

Claude Code (optional)

posts published

from drafts

attributed

$0.00

signups + revenue

Deploy this skill as a cloud automation to put it to work — runs on a schedule, dispatches output to Telegram or email, and logs every result here.

Deploy this skill

no laptop required

Run this skill on a schedule in the cloud. Pipe output to Telegram, email, or any webhook. Every run is logged here with full output, cost, and thumbs feedback.

How this performs

Cloud-automation runs, real outputs, and feedback.

No cloud runs yet for this skill.

Once you deploy it (above), every run will show up here with the actual output, exact cost, and a thumbs button so you can rate quality over time.

what this is

Claude Code skill at /Users/olanrewajukareem/.claude/skills/agent-sdk-patterns/SKILL.md. Auto-triggered by description matching when you type in Claude Code, OR embedded into agent system prompts (LeadFlow framework skills are wired this way). Auto-synced from disk every 6h via launchd.

Use this skill when building autonomous agents, multi-step LLM workflows, tool-using agents, or any logic that uses the Claude Agent SDK in Jay's projects. Triggers on "Claude Agent SDK", "agent loop", "tool use", "AI workflow", "autonomous agent", "

Local Claude Code use (optional)

Install the SKILL.md file locally so this skill is available the next time you open Claude Code on your Mac. Most users skip this and just use cloud deployments above.

Install on your machine

Save this skill to ~/.claude/skills/agent-sdk-patterns/SKILL.md so Claude Code auto-loads it whenever its description matches your prompt.

30-day activity

never used

2026-04-03today

cost model

unitincluded

capabilities

{
  "source": "global",
  "description": "Use this skill when building autonomous agents, multi-step LLM workflows, tool-using agents, or any logic that uses the Claude Agent SDK in Jay's projects. Triggers on \"Claude Agent SDK\", \"agent loop\", \"tool use\", \"AI workflow\", \"autonomous agent\", \"agent pipeline\", or \"multi-step LLM\". Applies Jay's patterns for tool definition, system prompts, retry logic, cost logging, and turn limits."
}

how it gets called

{
  "method": "embed_in_prompt",
  "skill_path": "/Users/olanrewajukareem/.claude/skills/agent-sdk-patterns/SKILL.md"
}

This tool has never been called in the last 30 days.

That means either: (a) the system has not had a task that matched its strengths, (b) a more cost-effective alternative was picked by the router, or (c) the related automation is toggled OFF on /automations. The Recommender reviews idle tools weekly and may suggest killing it.