The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes external commands to interact with the service backend. Evidence: The script _hws_cli.py uses subprocess.run() to invoke clouddocs-cli for operations like space fetching, document creation, and authentication status checks. Mitigation: The skill correctly passes arguments as a list rather than a single string, which prevents common shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of a binary tool on the user's system. Evidence: bin/windows/install-windows.ps1 automates the extraction and deployment of clouddocs-cli.exe to a local directory and modifies the user's PATH environment variable. Safety Measure: The installation script includes a SHA-256 integrity check against a provided SHA256SUMS file to ensure the binary has not been tampered with.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content which could potentially contain malicious instructions intended to mislead the LLM or influence the generated document. 1. Ingestion points: User input (topics/text) is ingested via SKILL.md and written to drafts/page.md. 2. Boundary markers: There are no explicit delimiters or instructions within the generated Markdown file to ignore embedded commands. 3. Capability inventory: The skill can write files (page.md, created_doc.json) and execute shell commands (clouddocs-cli). 4. Sanitization: Shell commands take inputs as list items, preventing direct command injection into the shell, but the resulting document content is not sanitized for downstream LLM consumers.

huawei-cloud-clouddocs-aipage