The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides setup instructions that involve downloading scripts from a remote domain and piping them directly into a shell interpreter (curl | bash and irm | iex). This pattern allows for the execution of unverified remote code on the host system without local audit or package verification.\n
Evidence in SKILL.md: curl -fsSL https://d.officecli.ai/install.sh | bash\n
Evidence in SKILL.md: irm https://d.officecli.ai/install.ps1 | iex\n- [COMMAND_EXECUTION]: The skill facilitates extensive file system interaction through the execution of shell commands using the officecli tool, which is used to create and modify Word documents.\n
Evidence in SKILL.md: The skill directs the agent to use verbs like create, add, and set to modify Word files.\n- [INDIRECT_PROMPT_INJECTION]: The skill reads data from external .docx files, which creates a potential vector for indirect prompt injection if the files contain malicious instructions meant for the AI agent.\n
Ingestion points: officecli view commands are used to read text and HTML from documents (described in SKILL.md).\n
Boundary markers: None are provided to prevent the agent from following instructions found within the document data.\n
Capability inventory: The skill includes diverse file modification capabilities as detailed across multiple scripts and instructions in SKILL.md.\n
Sanitization: While the skill includes instructions for escaping shell metacharacters, it does not include semantic content sanitization for the agent's logic.

officecli-docx