docs(browser): guide recovery from incomplete extraction

[doudouOUC]

Description

When extraction returns an empty body or only visible spreadsheet rows, an agent can keep trying selectors on the wrong page or report a partial result as complete. The browser skill now directs the agent to verify its named session's URL/title, recover the intended page, try AX for missing code-card text, and check full grid coverage through an API, export, or traversal. It also clarifies that an arbitrary DOM array position is not an OpenCLI interaction ref.

Related issue: Closes #2482. Complements #2478, which documents session visibility and binding. This is a documentation change based on a historical 1.6.1 transcript and a review of current main; it does not claim to reproduce or fix the old runtime session-loss behavior.

Type of Change

• 🐛 Bug fix
• ✨ New feature
• 🌐 New site adapter
• 📝 Documentation
• ♻️ Refactor
• 🔧 CI / build / tooling

Checklist

• I ran the checks relevant to this PR
• I updated tests or docs if needed
• I included output or screenshots when useful

Validation: the skill validator and git diff --check pass. The URL/title, AX, session, and numeric-target instructions were checked against current CLI registrations and the existing session contract. No runtime code changed; browser E2E, build, and unit suites were not run. The original logged-in environment was not available for reproduction.

Documentation (if adding/modifying an adapter)

N/A — no adapter changes.

• Added doc page under docs/adapters/ (if new adapter)
• Updated docs/adapters/index.md table (if new adapter)
• Updated sidebar in docs/.vitepress/config.mts (if new adapter)
• Updated README.md / README.zh-CN.md when command discoverability changed
• Used positional args for the command's primary subject unless a named flag is clearly better
• Normalized expected adapter failures to CliError subclasses instead of raw Error

Screenshots / Output

Skill is valid!
git diff --check: passed
1 file changed, 15 insertions(+), 1 deletion(-)

For review, consider an unexpected blank page, code blocks missing from DOM output, a sheet exposing only a few visible rows, and a DOM enumeration returning i=216. The guidance should lead to checking the target first, choosing an appropriate extraction fallback, reporting unverified coverage as partial, and obtaining a real interaction ref before acting. A screenshot or a fully consumed text cursor is explicitly not treated as proof of whole-sheet completeness.