Agent 使用體驗稽核

評估 Agent 使用文件或 SDK 時遇到的障礙。

Browserbase · 官方來源 · AI 與 Agents

來源狀態

可用來源

資源類型

Skill

與類似 Skill 有什麼不同？

GitHub Copilot SDK 應用整合
Agent Experience 著重讓產品或工具易於 Agent 使用；Copilot SDK 著重以 SDK 將 Agent 能力整合進應用。

來源描述的能力

這些是來源描述的可能操作，不代表 2lus 已授予權限或已測試。

尚未列出能力，請審查原始來源。

這個 Skill 是什麼？

比較精簡任務提示下的操作紀錄，找出文件與入門流程問題。

可以做什麼？何時適合使用？

SDK 入門審查
Skill 可用性
文件修訂比較

如何使用

指定目標、成功條件及成本上限。
先選擇 draft-only 或經授權的執行模式。
核對操作證據與結論，分開列出阻擋和未執行項目。

使用前你需要準備

Agent 使用情境
服務契約與授權範圍

你可以替換：

[agent]、[tool name]

環境與相依需求

具 subagent、WebFetch、Bash 能力及網路的執行環境；受測產品可能需要自己的 SDK、帳戶與密鑰。

使用範例與 Prompt

以下是 2lus 撰寫的示範需求；請替換為你有權處理的檔案與專案，不代表已執行或保證結果。

入門

只規劃公開 SDK 文件稽核，採 draft-only，不搜尋任何憑證。

實務

比較兩版快速入門的安裝及錯誤提示，列出有證據的摩擦點。

進階

規劃有預算上限的多 Agent 比較，限測試帳號，禁止憑證自動探索並保存去識別化紀錄。

實用提醒

固定任務與預算，比較前後文件版本。

限制與注意事項

小樣本評分不代表普遍相容；不同模型及網路會影響結果。

安全注意事項

預設採無憑證或 draft-only；不要允許自動搜尋私人設定檔，執行命令需明確授權。

使用與設定

此條目不提供已確認的通用安裝指令；請依官方文件與 Agent 版本操作。

支援平台

Claude Code — 文件記載，未做實機相容性測試官方文件 ↗

來源與授權

來源查核日期（非安全認證）: 2026-09-29

MIT

原始來源 ↗ 授權條款 ↗ 官方文件 ↗

Agent Experience

Evaluate obstacles agents encounter in documentation or SDKs.

Browserbase · Official source · AI & Agents

Source status

Active source

Resource type

Skill

How is this different from similar skills?

Copilot SDK
Agent Experience focuses on making products usable by agents; Copilot SDK integrates agent capabilities into applications through an SDK.

Documented capabilities

These are operations described upstream, not permissions granted or tested by 2lus.

Capabilities not declared here; review the original source.

What is this skill?

Compare traces from concise tasks to find onboarding and documentation problems.

Use cases and when to use it

SDK onboarding review
Skill usability
Documentation revision comparison

How to use it

Specify target, success criteria and cost ceiling.
Choose draft-only or an explicitly authorized execution mode.
Check evidence against conclusions; separate blocked and unexecuted work.

What you need

Agent use cases
Service contracts and authorized scope

You can replace:

[agent], [tool name]

Environment and dependencies

A harness with subagents, WebFetch, Bash and network access; target products may need their own SDKs, accounts and keys.

Usage and prompt examples

These example requests were written by 2lus. Substitute files and projects you may use; examples are not executed results or guarantees.

Beginner

Plan a public SDK documentation audit in draft-only mode without credential discovery.

Practical

Compare installation and error guidance in two quickstarts; list evidence-backed friction.

Advanced

Plan a budget-limited multi-agent comparison using test accounts, prohibit credential auto-discovery and retain redacted traces.

Tips

Keep tasks and budgets fixed across document revisions.

Limitations

Small-sample scores do not imply general compatibility; models/network affect results.

Security notes

Prefer no credentials or draft-only; prohibit private-file credential discovery and explicitly authorize commands.

Usage and setup

No verified universal installation command is provided for this entry. Follow the official documentation for your agent version.

Supported agents

Claude Code — Documented; not runtime-tested Documentation ↗

Sources and license

Source check date (not a safety certification): 2026-09-29

MIT

Original source ↗ License terms ↗ Documentation ↗

Related skills

Before using a third-party skill, review its source, permissions and executable content. Commands are for viewing and copying only; 2lus does not execute them.

2lus AI Skills Library provides curated educational guides. Third-party content, licenses and availability are governed by their original sources. Review permissions and executable content before use or installation.

Agent 使用體驗稽核