feat: support explicit page targets for browser actions

[254808127]

Summary

Add explicit page selection for multi-tab browser sessions:

• All 10 action subcommands accept optional --target-id, before or after the action subcommand.
• SDK callers can use Cdp::connect_to_target(ws_url, page_id).
• Missing/closed or non-page targets fail without selecting another page or creating a blank page. Attachment races propagate the CDP error.
• Document how an agent discovers a popup through existing session targets, selects its page ID, and uses it on subsequent commands.

This is an additive capability: the existing default selects the first page and cannot promise to follow a newly opened tab. Explicit target selection makes that choice independent of target enumeration order.

Example

browser-cli session targets --session-id SESSION_ID
browser-cli action snapshot --session-id SESSION_ID --target-id PAGE_ID
browser-cli action click --session-id SESSION_ID --target-id PAGE_ID --selector '#search'
# Re-list and choose the result page if the click opens a new tab.
browser-cli session targets --session-id SESSION_ID
browser-cli action snapshot --session-id SESSION_ID --target-id RESULT_PAGE_ID

Compatibility and scope

• Without the flag, existing behavior and action JSON output shapes remain unchanged.
• No implicit popup switching, persisted active-page state, click-semantics change, new dependency, or authentication change.
• Prepare version 1.2.0 in Cargo.toml, the root Cargo.lock package entry, and both platform bootstrap pins. No dependencies or installer behavior change beyond the pin.
• Skill instructions still check the actual binary's action --help; published 1.1.15 does not support the feature. Updating a Skill does not replace its existing executable. The pinned 1.2.0 assets must be published before installation/upgrade can succeed.
• No tag or release is created by this PR update. Publication remains a separate step with matching versioned assets and Skill package.
• No test-machine deployment or cloud/Harness evaluation results are included.

Local verification

Verified on Windows with isolated loopback HTTP/CDP fixtures and a separate headless Chromium profile:

• cargo fmt --all -- --check
• cargo test --all-targets --locked --offline — 27 passed; optional browser test ignored in the default suite.
• cargo clippy --all-targets --locked --offline -- -D warnings
• cargo test --doc --locked --offline
• cargo test --locked --offline --test page_targets_browser -- --ignored --nocapture with BROWSER_CLI_TEST_CHROME — 1 passed, HeadlessChrome 145.0.7632.6 / CDP 1.3.
• Skill frontmatter/reference validation, shell syntax, and whitespace checks.
• Rebuilt CLI reports 1.2.0, its action --help exposes --target-id, and Cargo/lock/both bootstrap versions agree. Dependency entries and the remaining installer logic are unchanged.
• 19 Windows PowerShell 5.1 bootstrap-selection regressions pass after synchronizing the merged CI fix.

Tests cover all action routes, reversed target order, invalid targets, target closure between listing/attachment, legacy defaults, output/file invariants, actual popup discovery, three independent CLI reconnects, unrelated tab lifecycle, result-page mutation, and closed-target failure without touching the source page.

macOS/Linux runtime and real cloud/WorkBuddy end-to-end behavior have not been locally verified. Existing CI can run the platform/build checks; this PR does not release the feature.

Self-review before this update

The implementer reviewed the complete feature diff and the version-update increment
before pushing: all action routes, default compatibility, missing/non-page/closed
targets, attachment races, fixture isolation, package/lock/pin consistency, and
release-versus-installation guidance. No blocking findings remain. Existing
capability checks are preserved, and the documentation explicitly distinguishes
Skill updates from executable upgrades and source merges from published assets.
This is self-review, not a claim of independent approval or cloud/Harness validation.

CI baseline

This branch now includes main's separately reviewed fix #29. The earlier
bootstrap failure
was caused by blindly selecting a missing historical release, not the page-target
feature. The CI workflow and selector match the updated main; no checks are skipped.

The bootstrap smoke test checks the newest usable already-published COS release,
while Rust tests/builds exercise this branch's 1.2.0 source. A green bootstrap check
does not claim that 1.2.0 has been published. The current results are shown in the
PR checks; publication and test-machine upgrades are outside this update.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-14T11:26:31.135564Z	a1dc32f	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Copilot]

🔵 Needs a closer look

The change spans CLI, SDK, browser behavior, tests, and documentation and warrants final human review.

Pull request overview

Adds explicit page-target selection for browser actions and SDK callers while preserving existing default behavior.

Changes:

• Adds --target-id support to action commands.
• Adds strict target attachment and validation.
• Adds protocol, CLI, browser, and documentation coverage.

File summaries

File	Description
tests/support/mod.rs	Shared isolated CLI test support
tests/page_targets.rs	Deterministic target-selection tests
tests/page_targets_browser.rs	Browser popup and lifecycle tests
src/main.rs	CLI parsing and action routing
src/cdp.rs	Target validation and attachment
skills/lexmount-browser/SKILL.md	Multi-tab workflow guidance
skills/lexmount-browser/references/troubleshooting.md	Target troubleshooting guidance
skills/lexmount-browser/references/commands.md	Command reference updates
README.md	User-facing documentation

Review details

• Files reviewed: 9/9 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.