Nature Literature Downloader

The skill demonstrates strong security awareness at the documentation level: it explicitly prohibits bypassing paywalls, DRM, or captchas; does not read or export cookies, passwords, localStorage, or session files; API keys are passed via stdin and stored with 0600 permissions, avoiding echo in command arguments, logs, replies, and manifests; requires SI confirmation to prevent unintended downloads; and includes provider-friendly pacing and audit trails. However, the core functionality relies on the user's logged-in browser and CDP proxy, involving operation on the user's browser environment, while permission control, session isolation, and rollback mechanisms are not fully specified; the docs mention limitations but lack implementation details such as the scope, restrictions, and failure recovery of browser control. Deductions: incomplete permission and isolation, and lack of recovery mechanisms.

The skill provides detailed workflows, status codes, and failure handling, but key scripts (e.g., batch_download.mjs, configure_school.py, browser_pdf_downloader.mjs) and their test suites were not reviewed, so correctness cannot be verified. The docs explicitly require Node 22+ and Python 3, but dependency lists and installation instructions are missing, which may cause environmental issues. Moreover, many failure statuses (e.g., verification_auto_failed, carsi_waiting_user) depend on browser state, and their implementation cannot be confirmed from static review. Deductions: static review only, tests not executed, and dependence on external services (CNKI, Web of Science) that may not be directly accessible from mainland China, potentially affecting reliability.

The skill has clear target audience (users needing lawful access to academic full texts) and scenarios (CNKI institutional access, OA retrieval, publisher APIs), but capability boundaries and trigger conditions are clear in the docs, yet actual triggering depends on user input and configuration. Chinese language support is good, but the core functionality depends on overseas services like CNKI and Web of Science, which may be restricted or require specific access methods in mainland China, affecting adaptability. Deductions: environmental fit and accessibility not verified, and reliance on overseas services may hinder use in mainland China.

The skill has a complete information architecture: SKILL.md, README.md, README_EN.md, references/, and data/. The docs provide configuration, installation, examples, and boundaries, but lack explicit known limitations (e.g., compatibility with specific schools) and version/changelog. The publisher is unverified, but the docs provide clear maintainer information, so responsibility is mostly clear. Deductions: missing known limitations and change history.

The skill claims to complete download tasks, but static review cannot verify actual output quality. The docs specify clear output formats (manifest.json, PDF, HTML, etc.) and detailed example commands, but lack independent verification of output usability. Deductions: static review cannot confirm whether outputs are directly usable, and dependence on browser state and external services may affect actual effectiveness.

The docs provide test commands (python3 -m unittest discover -s tests/python and node --test tests/unit/*.test.mjs), but test files and results were not reviewed, so authenticity cannot be confirmed. An internal design doc (docs/nature-downloader-reference-research-2026-07-15.md) provides external references, but no specific validation data is seen. Deductions: tests not executed, insufficient external validation data, only static documentation evidence.