web-reverse

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). This skill explicitly requires capturing tokens/cookies/headers and producing runnable replay scripts and token_flow artifacts that document key/token/nonce sources, which forces inclusion of secret values verbatim in outputs (commands/code/requests), creating exfiltration risk.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The skill's required workflow explicitly fetches and ingests public web pages and scripts (see SKILL.md steps like "抓当前题页 HTML 指纹 / 枚举当前加载的 script/link/API 资源") and the included runtime tools (web_replay.js, web_replay_guide2.js, solver.py) perform HTTP requests and evaluate page-sourced JS from the public site (e.g., match.yuanrenxue.cn), so untrusted third-party content is read and interpreted and can materially influence subsequent actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). Yes — the skill fetches and executes code from the target pages at runtime (web_replay.js requests and decodes/executes a payload from https://match.yuanrenxue.cn/match/1 via vm.runInContext, and solver.py loads and depends on runtime page scripts from https://match.yuanrenxue.cn/match/10 to obtain and invoke signer functions), so external content from those URLs directly controls execution.