Make public onboarding runnable and document dependency provenance

[lanyue-llk]

Changes

Align About, both READMEs, package metadata and architecture around concrete benchmarks: web fetching, trajectory replay and AI agent automation, comparing task completion, latency and resource use. Mark these as current tracks. Keep the Web Fetch CLI help specific to its actual scope.

New users could run the offline fixture, but the README did not explain which browsers, methods and benchmarks were available or provide a runnable real-adapter comparison. It also described Python 3.11 as an exclusive requirement despite broader base-package support.

• Restructure English/Chinese READMEs around choosing a track, running an example, reading results and publishing. Fold detailed participant/dataset lists, move full local-browser instructions into their example guide, and simplify repository navigation. Rewrite setup, metrics, hosting, contribution and dependency prose; add task-oriented navigation and shorter sections to the technical references. Keep scoring, qualification and reproduction requirements. Remove an unsupported future-backend proposal. Add an owned local article and a compact configuration for HTTP, Moli, Chromium, Lightpanda and Obscura, using the existing runner and scorer.
• Test the base CLI on Python 3.12, 3.13 and 3.14 in CI while retaining the pinned Python 3.11 development and historical experiment environment. Explain interpreter selection for both installation and execution.
• Present browsable, evidence-linked reports and Artifact Site publishing in both READMEs. Add China/global registration entries, a repository navigation table covering the companion browser/protocol benchmark, maintained benchmark forks and Agent baselines. Document optional Artifact Site registration/authorization and local report hosting. Use checked CLI 0.4.0 commands with --share none; distinguish diagnostic uploads from profile-bound certified releases and their access controls.
• Document dependency repositories, full commits, relevant branches and lockfiles. Remove the macOS-only Chromium path from .env.example. Select Python 3.11.12 in the historical OSS bootstrap to match its declared library-adapter identities.
• Remove the historical version from the unversioned moli display label. Actual browser versions remain in execution evidence. No scoring or historical evidence changes.
• Vendor Humanizer 3.1.0 from upstream commit 225a6f39ac85f76ee48dbad772ea4abe4ed6c9d8. Humanizer Chinese remains at upstream HEAD 1d6c08e10fbfe0e1b36ac784834451c6aae0a92c / 1.0.0. Retain upstream skill/license bytes unchanged and add provenance, update instructions and integrity checks.
• Add CITATION.cff and matching README citation examples for v0.1.0, with repository URL and separate upstream attribution guidance.

Validation

• Positioning follow-up: 36 public-entrypoint/package tests and 6 subtests passed; uv lock --check --offline passed. Checked actual CLI help and verified the GitHub About description after updating it.
• Documentation follow-up: 43 public CLI/scenario tests and 6 subtests passed. All 154 local links/anchors across the 17 edited Markdown files resolve; README command/citation blocks and repository links match across languages. Local Chromium previews exercise folding tables, section navigation and desktop/narrow layouts in both languages. No runtime code, datasets or scoring rules changed in this follow-up.
• China/global hosted entry points and deployment-specific publishing guides returned HTTP 200. Validated 66 local links/anchors and matching README commands/citations; kept the global /artifact-site prefix in publishing instructions.
• CI passed at b13a1f5: the full repository job and base CLI jobs for Python 3.12, 3.13 and 3.14.
• 93 relevant tests and 11 subtests passed, covering public entry points, report labels, acceptance, saved evidence, package contents and vendored source integrity.
• Base CLI and local HTTP scenarios passed on Python 3.12, 3.13 and 3.14. Complete content passes; a missing final paragraph fails. Reports rebuild after the local server stops without further provider calls.
• Ran a real HTTP/Moli/Chromium matrix against the owned article: all three passed and produced one report. Rebuilt and opened the home, analysis and Moli result detail in Chromium with no page errors and no invented historical browser version. Ran LangChain WebBase and Crawl4AI against the same article using the pinned Python 3.11.12 OSS environment; both passed and generated a report.
• Copied skills and licenses match downloaded pinned upstream bytes. Fixed Git dependency commits exist; the pinned WebArena evaluator is publicly accessible without GitHub authentication.
• CFF schema 1.2.0 validation, Markdown link/command parity checks, Ruff and track naming checks passed. Frozen acceptance hashes remain unchanged.

Follow-up to #44, which is already merged. The existing v0.1.0 tag and historical measurements are unchanged. Initial public distribution still requires a current-source export rather than exposing the old development history.

[lanyue-llk]

审查结论：通过。本轮未发现需修改的新增问题。

PR 45 补齐了公开仓库的入门路径：真实适配器示例、依赖来源、Python 兼容性、报告托管和引用信息。示例通过配置复用现有运行器、评分器与报告构建器，方案够用，没有重复实现，也不需要再加一层抽象。此前 PR 44 最后指出的 macOS 系统路径别名问题已在合入版本修复。

验证结果：

• 本地实跑 HTTP、Moli、Chromium、Lightpanda、Obscura：完整文章全部通过，删除末段后全部失败；关闭页面服务后，两组证据均能重建报告。
• 全量 Python 回归：5081 项测试、1711 项子测试通过，55 项跳过，无失败；报告 UI、加载器与 Browser Use UI 的 Node 检查通过。
• 当前提交的 CI 全部通过，包括 Python 3.12、3.13、3.14 基础 CLI。
• Humanizer 和 Humanizer Chinese 的技能及许可证字节与固定上游提交一致；依赖版本与仓库绑定相符；托管命令已对照 CLI 0.4.0 核实。