The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.Popen within its Python scripts (run_with_env.py, scan_drive.py) to interface with the aliyunpan command-line utility. This is the intended primary purpose of the skill. The execution is restricted via an ALLOWED_EXECUTABLES whitelist that only permits aliyunpan or aliyunpan.exe, preventing arbitrary command injection.
[EXTERNAL_DOWNLOADS]: The documentation references official installation methods such as Homebrew (brew install aliyunpan) for the Aliyunpan CLI. This is a well-known and trusted package manager.
[CREDENTIALS_UNSAFE]: The skill implements a strict 'no-log' policy for tokens and cookies. It uses a dedicated wrapper run_with_env.py to load sensitive environment variables from a private configuration file (~/.config/soia-skills/.../config.yml) directly into the sub-process, ensuring credentials are never printed to logs or standard output.
[PROMPT_INJECTION]: No evidence of system prompt overrides or safety bypass attempts was found. The instructions clearly state that agents must verify user requests for destructive actions like rm.
[DATA_EXFILTRATION]: Network operations are limited to those performed by the official aliyunpan CLI directed at Aliyunpan's API. No unauthorized data exfiltration patterns to third-party domains were detected.

soia-pkm-alipan-drive-ops