| name | web-capture |
| description | Capture public web pages into the Second Brain as clippings using a real browser automation tool (Chrome DevTools MCP preferred when available). Use when a user provides a URL and wants it clipped/summarized/routed with sources, especially for JS-heavy sites. |
web-capture(Public Web → Clipping)
Purpose
Turn a public URL into a vault clipping with:
- link
- short summary
- key excerpts (optional)
- source links
- next actions (optional)
This skill is optimized for:
- JS-heavy sites
- pages that render content dynamically
- cases where simple HTML fetch fails
Safety / boundaries (hard)
-
Public pages only by default.
- If the page requires login, paywall, or contains personal/private content → STOP and ask for explicit permission + scope.
-
Operator confirmation shorthand (allowed)
- If the user replies with 「確認」 for a given URL request, treat it as:
- public page, no login
- no personal/private data
- ok to capture and quote
- If the page visibly shows a login wall, private profile, or requires account access, STOP anyway and report the block (do not proceed).
-
Do not bypass access controls or “break” protections.
-
Always preserve source attribution (URL + publisher).
Tooling strategy (in order)
- Chrome DevTools MCP (preferred)
- Use when you have a connected Chrome session and need reliable rendering/interaction.
- Simple fetch
- If the content is static and fetchable, prefer a lightweight HTML fetch to save cost.
If the runtime does not have Chrome MCP, use the best available browser tool.
Output contract (always)
Minimal prompt template (for human operators)
When calling an agent:
- URL:
- Capture goal: (summary only | summary + key quotes | structured extraction)
- Destination: (11_Clipping | project name)
- Constraints: (language, length, citations)
References
- For folder destinations: read
references/folder-map.md.
- For clipping schema: follow your vault protocol.