docs: prepare 0.3.0 guidance and streamline browser skills

[iuyo5678]

Summary

Prepare the documentation for 0.3.0 by closing gaps between installation, skill discovery, first use, and upgrades. Shorten the bundled browser skills while preserving concrete examples and recovery guidance for agents.

This PR changes 12 Markdown files.

Changes

Installation and release guidance

• Make the installed CLI available in the current Unix shell and explain PATH handling in running agents.
• Select the intended harness explicitly, distinguish the DSH plugin setup, and verify skill discovery separately from bsk doctor. A successful first-use check must read a page and stop its session.
• Document CLI, daemon, extension, and plugin upgrades, including custom daemon settings and the 0.3.0 migration to browser-controlled Automation settings.
• Align English and Chinese setup/privacy guidance with current permissions, Quick Actions, remote tab ownership, file-transfer limits, and audit storage on the selected daemon host.
• Complete the Unreleased changelog with the existing features and behavior changes relevant to 0.3.0.

Skill compression

• Organize instructions around the task lifecycle, common interactions, and recovery; replace repeated explanations and the full command inventory with targeted help lookup.
• Keep usable examples for session handling, forms, borrowing, screenshots, and file transfer. Preserve the rules for stale refs, unknown effects, human help, Canvas captures, and cleanup.
• Clarify that ending a session returns borrowed tabs while leaving those tabs open in the user's window.
• Keep single-file distribution and identical canonical/packaged CLI skills. The DSH skill continues to use its six native tools and loaded parameter schemas.

Full SKILL.md token counts using o200k_base, compared with this PR's main base:

Skill	Before	After	Reduction
CLI skill	4,206	2,959	29.6%
DSH native-tool skill	1,459	1,304	10.6%

These counts cover instruction text, not total tokens consumed by a browser task.

Validation

• cargo test -p bsk --locked --test install_skill --test cli_parse — 45 passed.
• cargo test -p bsk --lib --locked skill_install — 38 passed.
• Regenerated the DSH skill content and ran tests/skill.test.ts — 5 passed.
• Parsed all 26 complete CLI examples with the current CLI argument parser.
• Validated skill frontmatter, checked canonical/packaged CLI skill equality, and passed git diff --check.
• Completed 14 decision scenarios covering continuation refs, hover menus, upload uncertainty, disabled help, Canvas capture validity, rejected borrowing, and session cleanup. Rechecked the session/tab lifetime clarification after the scenario review.

Real-browser smoke tests

An independent agent context loaded the committed CLI skill and completed these tasks through the actual browser:

Website	Task and observed result	CLI calls
Wikipedia	Used site search for Ada Lovelace and read the full birth/death dates from the article.	8
Selenium official demo	Filled synthetic text values, resolved the dropdown's actual option value, selected Two, and submitted the demo form. The page showed “Form submitted” and “Received!”.	12
Python documentation	Used site search for pathlib, opened its reference page, read the introduction, and saved and inspected a viewport screenshot.	10

All three tasks passed: 30 CLI calls, zero command failures, and all three sessions stopped. A final status check found zero active sessions. Calls include one command-help lookup.

The installed CLI, daemon, and extension reported version 0.2.1; the daemon used protocol 1.1 and the extension used protocol 1.3. These smoke tests cover the common CLI flows above. They do not establish a smaller-model success-rate comparison or DSH native-tool end-to-end coverage. Large observations produced truncation warnings; the required controls and evidence remained available after applying explicit token limits.