引導式測驗 · 硬 等級
提示市場與公開提示集合小測驗
測試 LangSmith 公開的信任邊界、諮詢影響和修補版本,以及安全審查、固定和憑證實踐。
問題 1 的 8
What does LangChain say about prompts in its public LangSmith hub?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- What does LangChain say about prompts in its public LangSmith hub?
- Published prompts have verified behavior for all model versions
- They are community-created, user-generated, unverified, and not reviewed or endorsed
- The hub’s public entries are reviewed and endorsed before publication
- Every public entry is tested against each user’s application
- Why can an SDK pull involve more risk than copying visible prompt text?
- A public SDK pull can return only plain text and cannot deserialize configuration
- Copying prompt text into local code follows the same manifest-deserialization path as an SDK pull
- The risk appears only when the model is retrained on a public prompt
- The manifest may contain serialized objects and model configuration that affect runtime behavior
- Which impact does GHSA-3644-q5cj-c5c7 describe?
- A malicious manifest can redirect model requests and potentially expose prompt data or credentials
- Viewing the public page automatically grants its author workspace administrator rights
- A pull changes the trained weights of the model whether or not it is used
- Every visit to the hub automatically sends local credentials to the prompt author
- Which Python langsmith version is listed as the patched floor in the advisory?
- 0.3.30, the patched langchain floor
- 1.0.7, the patched langchain-classic floor
- 0.8.0
- 0.6.0, the patched JS/TypeScript langsmith floor
- Which JavaScript/TypeScript langsmith version is listed as patched?
- 1.0.7, the patched langchain-classic floor
- 0.6.0
- 0.8.0, the patched Python langsmith floor
- 0.3.30, the patched langchain floor
- What does the patched SDK require before pulling a public owner/name prompt?
- An explicit public-prompt opt-in acknowledging the trust boundary
- The SDK silently loads public prompts and warns only after deserialization
- The public-pull opt-in applies only to private prompts from the same organization
- Upgrading removes model configuration from every manifest, so review is unnecessary
- Which practice reduces reliance on a mutable public prompt reference?
- Use the owner/name reference alone because it permanently fixes prompt contents
- Review the current branch once and keep following its moving head in production
- Trust the publisher account without inspecting the prompt version
- Review and pin the exact approved commit
- What should teams avoid when using an untrusted external manifest?
- Set include_model=True to expand the deserialization allowlist for unknown models
- Use broad credentials because the SDK now blocks public pulls by default
- Broad model deserialization and loading secrets from the environment
- Set secrets_from_env=True so the prompt can resolve any needed variables