The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file describes a workflow where the check_skill_update tool returns a JSON object containing an instruction field. The skill directs the agent to "follow the instruction command to update" (遵循 instruction 指令进行更新). This pattern involves executing dynamically generated strings from a remote server, which is a significant security risk if the agent automates the process without user validation.
[COMMAND_EXECUTION]: The setup.sh script executes several shell commands to manage the environment, including installing global Node.js packages (npm install -g mcporter) and modifying tool configurations. It also generates and stores authentication tokens in the user's temporary directory.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It ingests untrusted data from survey participants via the list_answers tool.
Ingestion points: The list_answers tool (in references/list_answers.md) and get_survey tool (in references/get_survey.md) ingest user-submitted responses and survey content.
Boundary markers: None identified in the instructions to prevent the agent from obeying instructions embedded in survey data.
Capability inventory: The skill has the ability to execute CLI commands via mcporter and run the setup.sh script.
Sanitization: The documentation for get_survey notes that titles may contain HTML tags requiring cleaning, but there is no broader strategy for sanitizing instructions embedded in text fields.
[EXTERNAL_DOWNLOADS]: The skill relies on the external mcporter CLI tool and interacts with API endpoints at wj.qq.com (Tencent Survey). While these are established services, they represent external dependencies required for the skill's functionality.

tencent-survey