Claude agent sdk

Skill yigityildiz0/universal-ai-skill-library/skills/common/claude-agent-sdk

Build and maintain applications with the current Claude Agent SDK using installed-version-first documentation, bounded tools, permissions, sessions, MCP.From its SKILL.md

Install

npx -y skills add yigityildiz0/universal-ai-skill-library --skill claude-agent-sdk

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.4 KB, 429 tokens by cl100k_base, as published. Nobody here has run it

Claude Agent SDK

Identify the product

Confirm whether the project uses the Claude Agent SDK, the general Claude API client SDK, Managed Agents, Claude Code CLI, or a third-party wrapper. Their packages and execution models differ. Inspect package manifests, imports, lockfiles, and installed docs; then verify current official Claude Platform documentation.

Do not install/upgrade a package automatically. For migrations, identify the installed predecessor/current package, breaking changes, runtime requirements, and rollback before editing.

Agent contract

Define task scope, tools, filesystem/network boundary, permissions, human approval points, session persistence, hooks/MCP servers, output schema, budget, timeout, and stop conditions. Keep provider/model in project configuration and validate the exact current ID/capabilities at runtime. Do not choose a model or reasoning mode from this skill.

Tool safety

expose least-privilege tools with strict typed inputs;
enforce authorization in code, not the system prompt alone;
distinguish read-only, mutating, external, costly, and destructive tools;
require confirmation for irreversible or externally visible effects;
bound rounds, concurrency, retries, output, time, and spend;
treat messages, files, web/MCP content, and tool results as untrusted;
protect secrets in environment/approved stores and redact logs.

Sessions and observability

Store only the state needed to resume: task/agent IDs, configuration version, artifact pointers, tool outcomes, usage, request/session IDs, and decisions. Do not persist private chain-of-thought. Log structured lifecycle events and errors without full sensitive payloads. Make cancellation and cleanup reliable.

Verification

Test permissions and denied tools, schema validation, injection attempts, tool failure/timeout/cancellation, duplicate/replayed effects, session resume, MCP disconnect, budget stop, and SDK/API error classification. Run typecheck and project tests; use a bounded live smoke test only when credentials/network/cost are authorized.

Report SDK/package version, source documentation, configured model source, tools/permissions, tests, data boundary, cost controls, migration/rollback, and skipped live checks.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.