The Agent Skills Directory

[Remote Code Execution] (CRITICAL): In references/mcp-setup.md, the skill instructs the user to execute a remote script directly into the shell using curl -fsSL ... | bash. This is a high-risk pattern that allows for arbitrary code execution from a remote source that is not within the provided trusted scope.
[External Downloads] (MEDIUM): The skill relies on the octagon-mcp package via npx. This is an external dependency from an unverified source (OctagonAI) which is executed at runtime to facilitate the MCP server connection.
[Indirect Prompt Injection] (MEDIUM): The SKILL.md file defines a workflow where user-controlled parameters such as <INDUSTRY> and <EXCHANGE> are interpolated directly into a natural language prompt passed to the octagon-agent tool.
Ingestion points: User parameters defined in SKILL.md and README.md are used to build the tool argument.
Boundary markers: Absent. No delimiters or instructions are used to separate user data from the system's command instructions.
Capability inventory: The target octagon-agent tool has broad capabilities including web search and SEC filing retrieval, which could be abused if the prompt is manipulated.
Sanitization: None. The input is treated as trusted and directly concatenated into the tool's execution prompt.

industry-pe-ratios