[Bella-rab]

Summary

state show reports the wrong summary in both output modes.

The daemon built summary as a preformatted string ("12 cookies, 3 origins"), but the CLI read summary.cookies and summary.origins as numbers. Those lookups always returned None, so every state file printed Cookies: 0 and Origins with localStorage: 0 regardless of its contents.

While verifying, the text path turned out to be wrong for a second reason. A state-show payload carries filename and path, and the download formatter claims any payload with those fields. It runs earlier in print_primary_response, so state show printed a completed-download line and never reached the summary at all:

$ agent-browser state show my-session-default.json
✓ Downloaded to /home/user/.agent-browser/sessions/my-session-default.json (my-session-default.json)

Changes

• cli/src/native/state.rs: summary is now an object with cookies and origins counts. That matches the shape doctor already uses and what the CLI was written to read.
• cli/src/output.rs: the summary is extracted into format_state_show_text, and the state_show branch moves ahead of the generic path/filename branches. A preformatted string summary is still rendered, so a CLI pointed at an older daemon keeps working.

Verification

Built the binary and ran it against a fixture holding 2 cookies and 1 origin.

Before:

✓ Downloaded to ...\fixture.json (fixture.json)

After:

State file summary:
  Cookies: 2
  Origins with localStorage: 1

--json now returns "summary":{"cookies":2,"origins":1}.

Tests:

• cargo test --bin agent-browser format_state_show (5 passed), covering count rendering, the encrypted marker, a legacy string summary, the non-state payload, and the payload that used to match the download branch.
• cargo test --bin agent-browser state:: (19 passed), including a new daemon-side test that keeps summary structured.
• cargo test --bin agent-browser: 1918 passed, 4 failed. The 4 failures are pre-existing on Windows and unrelated to this change: install::tests::download_bytes_connection_refused_includes_details and session_settings::tests::* assert a localized error string and /path separators, and native::tab_binding::tests::test_pinned_owner_dead_or_missing_pid_is_free shells out to ps. The same 4 failed before this change.
• cargo fmt -- --check is clean.

Why the contract changed on the daemon side

Fixing only the CLI would have left --json consumers reading a sentence where they expect counts. The object shape is the one the CLI already assumed and the one doctor returns, so the daemon is the side that drifted.

Note

I did not find an open issue for this specific report. #2455 is a different state list problem (a blank modification date) and is already claimed by #2457.

Follow-up commit

74c31ee moves the new daemon-side test onto tempfile::tempdir(), matching the other tests in that module. process::id() is constant within the test binary, so the previous fixed filename could collide between tests and survive a panic.

[vercel]

@Bella-rab is attempting to deploy a commit to the Vercel Labs Team on Vercel.

A member of the Team first needs to authorize it.