id	web-content-fetch
name	Web Content Fetch
description	Fetch full page content when snippets are insufficient
role	product
requiredTools	[{"type":"builtin","tool":"fetch_web"}]

When snippets from web search are not enough:

Use fetch_web to retrieve the full content of a URL when the user or a prior search returned a URL that needs to be read in full.
Cite the URL in your answer; do not invent content that was not in the fetched page.
Use fetch_web when the user asks to "read this page", "what does this article say", or when search snippets are incomplete for the answer.
Handle errors and paywalls: if the fetch fails or returns a paywall, say so and base the answer only on what was returned (or that the content could not be retrieved).
Prefer fetch_web after search_web when you have a specific URL and need full-page extraction.

When the user provides a URL or search returned a relevant URL: call fetch_web with that URL to get full page content.
Parse the returned content for the information the user needs (e.g. main points, dates, data); cite the URL.
If the user asked "what does this say" or "summarize this page", provide a short summary with the URL.
If fetch fails (timeout, 403, paywall): report that the content could not be retrieved and suggest the user open the URL directly or try another source.
Do not invent or hallucinate content; only state what is present in the fetched result.

Input: "What does https://example.com/article say about X?"
Output: Short summary of the article's points about X from fetch_web result; cite the URL.
Input: "Search for recent news on Y and read the first result."
Output: Use search_web first; take the first result URL; call fetch_web on it; summarize the page and cite the URL.

Fetch fails: Say "I couldn't retrieve the page (timeout/error/paywall)" and suggest opening the URL in a browser.
Empty or minimal content: Some pages return little text (e.g. JS-heavy); report what was returned and that the page may need to be viewed in a browser.
Paywalled content: Base the answer only on any non-paywalled part returned; do not invent paywalled content.
User gives no URL: If the question implies a URL (e.g. "read the first result"), use search first to get a URL, then fetch.

fetch_web: Use when you have a URL and need full page content for summarization or extraction. Call with the URL; cite the URL in the answer; do not invent content.

id	web-content-fetch
name	Web Content Fetch
description	Fetch full page content when snippets are insufficient
role	product
requiredTools	[{"type":"builtin","tool":"fetch_web"}]

When snippets from web search are not enough:

Use fetch_web to retrieve the full content of a URL when the user or a prior search returned a URL that needs to be read in full.
Cite the URL in your answer; do not invent content that was not in the fetched page.
Use fetch_web when the user asks to "read this page", "what does this article say", or when search snippets are incomplete for the answer.
Handle errors and paywalls: if the fetch fails or returns a paywall, say so and base the answer only on what was returned (or that the content could not be retrieved).
Prefer fetch_web after search_web when you have a specific URL and need full-page extraction.

When the user provides a URL or search returned a relevant URL: call fetch_web with that URL to get full page content.
Parse the returned content for the information the user needs (e.g. main points, dates, data); cite the URL.
If the user asked "what does this say" or "summarize this page", provide a short summary with the URL.
If fetch fails (timeout, 403, paywall): report that the content could not be retrieved and suggest the user open the URL directly or try another source.
Do not invent or hallucinate content; only state what is present in the fetched result.

Input: "What does https://example.com/article say about X?"
Output: Short summary of the article's points about X from fetch_web result; cite the URL.
Input: "Search for recent news on Y and read the first result."
Output: Use search_web first; take the first result URL; call fetch_web on it; summarize the page and cite the URL.

Fetch fails: Say "I couldn't retrieve the page (timeout/error/paywall)" and suggest opening the URL in a browser.
Empty or minimal content: Some pages return little text (e.g. JS-heavy); report what was returned and that the page may need to be viewed in a browser.
Paywalled content: Base the answer only on any non-paywalled part returned; do not invent paywalled content.
User gives no URL: If the question implies a URL (e.g. "read the first result"), use search first to get a URL, then fetch.

fetch_web: Use when you have a URL and need full page content for summarization or extraction. Call with the URL; cite the URL in the answer; do not invent content.

web-content-fetch

المزيد من هذا المستودع