browser-live-click-content

Invoke keeper_compose_browser-live-click-content with the observed live
clientId, tabId, documentId, nodeId and current expectedUrl. This is a
callable composition, not an instruction to load through keeper_skill.

The follow validates the actual anchor and navigates directly to its href in
that tab; it does not execute page click handlers. New-window targets,
downloads and non-HTTP(S) destinations are rejected before navigation. If the
task needs a site's click handler, use the observed control's ordinary click
action and inspect its result before deciding what to read.

The successor reads visible content at the follow receipt's destinationUrl,
preserving its navigationSource. Use the returned body when it answers the
request. When a region must be selected first, choose
keeper_compose_browser-live-click-regions instead. Do not invoke both by
convention. A matching URL does not establish site readiness or complete
history: the site's instruction determines whether title, target and content
match the request, and whether another scoped observation is needed.

If the follow succeeded but the read failed, retain the follow receipt and
retry only BrowserRead on its pinned client/tab. Preserve navigationSource
on every retry. For the same destination URL, preserve expectedUrl too: a
same-URL reload must expose a new document identity. To inspect a possible
redirect, omit only expectedUrl and verify the observed destination and
content before adopting its URL. An unchanged source URL/document is still
pending; an authentication page or unrelated destination is unverified.
Never replay the follow merely because observation failed.

[[compositions]]
name = "browser-live-click-content"
description = "Follow one observed live anchor and return URL-guarded visible content. Retry only the read after an observation failure; site content still needs verification."
execution = "inline"

[[compositions.params]]
name = "clientId"
type = "string"
description = "Observed live browser connection UUID."

[[compositions.params]]
name = "tabId"
type = "integer"
description = "Observed live browser tab."

[[compositions.params]]
name = "documentId"
type = "string"
description = "Observed document identity of the chosen link."

[[compositions.params]]
name = "nodeId"
type = "string"
description = "Observed same-tab anchor identity, never a guessed selector."

[[compositions.params]]
name = "expectedUrl"
type = "string"
description = "Current observed URL before following the link."

[[compositions.nodes]]
id = "click"
tool = "BrowserInteract"
input = { kind = "object", fields = [
  { name = "lane", value = { kind = "literal", value = "live" } },
  { name = "action", value = { kind = "literal", value = "follow_link" } },
  { name = "clientId", value = { kind = "param", name = "clientId" } },
  { name = "tabId", value = { kind = "param", name = "tabId" } },
  { name = "documentId", value = { kind = "param", name = "documentId" } },
  { name = "nodeId", value = { kind = "param", name = "nodeId" } },
  { name = "expectedUrl", value = { kind = "param", name = "expectedUrl" } }
] }

[[compositions.nodes]]
id = "content"
tool = "BrowserRead"
after = ["click"]
input = { kind = "object", fields = [
  { name = "lane", value = { kind = "literal", value = "live" } },
  { name = "clientId", value = { kind = "param", name = "clientId" } },
  { name = "tabId", value = { kind = "output", node = "click", pointer = "/tabId" } },
  { name = "mode", value = { kind = "literal", value = "scene" } },
  { name = "expectedUrl", value = { kind = "output", node = "click", pointer = "/destinationUrl" } },
  { name = "navigationSource", value = { kind = "output", node = "click", pointer = "/navigationSource" } }
] }