The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes curl to interact with local API endpoints and python to perform system-level operations such as retrieving process IDs.
[DYNAMIC_EXECUTION]: Employs the python -c pattern to dynamically execute code strings at runtime to obtain the Parent Process ID (PPID) during the upload and logging workflow.
[DATA_EXFILTRATION]: Facilitates transferring local files to an external service (Tencent's 3g.qq.com). The skill accepts arbitrary absolute paths in the localPath parameter without validation or restriction, which could allow an agent to read and exfiltrate sensitive files, such as SSH keys, credentials, or environment configuration, if prompted to do so.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to tool output poisoning because it ingests a message field from a local API and strictly prohibits the agent from sanitizing, filtering, or modifying the content before displaying it to the user. (Evidence: Ingestion point is the API message field in SKILL.md; boundary markers are absent; capabilities include arbitrary file access via curl; and sanitization is explicitly forbidden by the instructions to output text verbatim).
[PROMPT_INJECTION]: The instructions use strong imperative language (e.g., 'CRITICAL', 'MUST', 'NEVER', 'Strictly prohibited') to override the agent's natural reasoning and safety filters, forcing it to act as a verbatim relay for server-provided content and concealing raw metadata from the user's view.

cloud-upload-backup