The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation provides an execute_skill_script function designed for AI agents to invoke helper scripts at runtime. This function constructs file paths using string interpolation (f'skills/{skill_name}/scripts/{script_name}') and executes them via subprocess.run. This pattern creates a potential risk for path traversal or unauthorized code execution if an attacker can influence the skill or script names.
[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke the Python interpreter on local script files. It also instructs users to run shell commands for installation (npx skills add, git clone).
[EXTERNAL_DOWNLOADS]: The skill is configured to download components and procedural definitions from a GitHub repository (mukul975/Anthropic-Cybersecurity-Skills) using Git and NPX tools.
[INDIRECT_PROMPT_INJECTION]: As a security analysis tool, the skill is designed to ingest and process potentially malicious data (malware samples, incident logs, memory dumps) which could contain hidden instructions for the agent.
Ingestion points: handle_query function processes natural language queries; forensics scripts ingest external files such as memory dumps.
Boundary markers: None identified in the provided templates to isolate processed data from agent instructions.
Capability inventory: The skill possesses capabilities for file system traversal (os.listdir), file reading (open), and command execution (subprocess.run).
Sanitization: There is no evidence of sanitization for arguments passed from user-provided kwargs directly to the subprocess.run command line.

anthropic-cybersecurity-skills