Tp4
- Category
- MCP Tool Poisoning
- Confidence
- 94% confidence
- Finding
声明描述的是一个功能完整的项目流程编排/交付技能,而实际代码只是一个配套日志工具。虽然日志记录可被视为“阶段追踪”的辅助实现细节,但该代码既不是所述系统的主体,也没有实现描述中的核心能力。其实际命令集与声明的命令集明显不一致,主要用途也缩减为日志写入与查看,因此属于明显的描述与行为不匹配。
- Content
Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed, project-local workflow helper that writes planning documents, logs, and status files but shows no evidence of deception, exfiltration, or unsafe persistence.
Install only if you want a Chinese-language project workflow skill that can modify code and create repository-local docs, logs, status markers, and .gitignore entries. Prefer explicit proflow commands, review generated files before committing, and avoid passing unusual custom IDs directly to helper scripts.
声明描述的是一个功能完整的项目流程编排/交付技能,而实际代码只是一个配套日志工具。虽然日志记录可被视为“阶段追踪”的辅助实现细节,但该代码既不是所述系统的主体,也没有实现描述中的核心能力。其实际命令集与声明的命令集明显不一致,主要用途也缩减为日志写入与查看,因此属于明显的描述与行为不匹配。
The declared description presents a broad project workflow orchestration skill with multiple high-level capabilities across the entire development lifecycle. The supplied code only implements a status manager: it tracks whether four stages are done, stores optional output path and requirement ID metadata, lists status, and resets markers. While stage tracking/reset is one subset of the declared functionality, the code does not implement the other core advertised behaviors such as brainstorming/planning execution flows, automatic feature-size evaluation, direct code path skipping, unique requirement ID management, document standardization/archiving, or broader command orchestration. Therefore the actual code materially under-delivers relative to the declared purpose, making the description inaccurate for this code chunk.
The skill explicitly allows 'fix' and '小功能' requests to skip all planning stages and proceed directly to locating and modifying code, yet that high-risk shortcut is not surfaced as a clear warning to users. In practice, this can lead to unreviewed code edits from vague prompts, increasing the risk of accidental breakage, policy bypass, or unsafe modifications triggered by misclassification.
The skill clearly instructs the agent to read and write repository files, create status markers, create .gitignore entries, and archive outputs, yet it declares no explicit tool scope or permissions boundary. In an agent environment, that omission can cause reviewers and users to underestimate filesystem side effects and may allow broader-than-expected file modification behavior.
The trigger conditions combine explicit commands with broad natural-language phrases like '项目全流程' and '一键全自动项目文档', which can cause the skill to activate in contexts where the user did not intend repository mutation. Because this skill is designed to create files, logs, and state artifacts, accidental invocation materially increases the chance of unintended changes.
All user-facing natural-language instructions in the skill are written in Chinese, and the file does not indicate that language is selectable or that the skill is intentionally limited to a Chinese-speaking environment. Under the policy, forcing a specific language without opt-in is a natural-language policy concern.
The skill states it will enforce directory conventions, generate IDs, initialize logs, and create status files, but it does not prominently warn users that it will modify repository contents automatically. Hidden or under-disclosed write behavior is dangerous in agent workflows because users may invoke the skill expecting advice, not filesystem changes and tracking artifacts.
The log file path incorporates the user-controlled requirement_id directly into the filename without validation or normalization. An attacker can supply path traversal sequences such as '../' to escape docs/logs and append to arbitrary files the process can access, which can corrupt files, poison logs, or interfere with other project artifacts.
This code file contains user-facing natural language in Chinese in the module docstring and function docstring, indicating a language-specific experience without any opt-in or alternative locale. Under the stated policy, language constraints should either offer user choice or be clearly justified as region-specific.
The module docstring is written entirely in Chinese and describes the skill's behavior in that locale, with no indication that users can choose another language or that the tool is region-specific. Under the policy rule for natural-language constraints, this can be treated as a locale/language preference imposed without explicit opt-in.
The file contains multiple user-facing descriptions and help strings in Chinese, including the module docstring and CLI help text, without indicating that the user can choose a language or that the tool is intentionally region-specific. Under the language/locale policy category, forcing a specific language without opt-in is a natural-language policy concern.