The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied information concerning vendor profiles and project risks to generate structured reports and registers. While this constitutes a data ingestion surface, the risk is mitigated by the skill's adherence to professional evaluation frameworks and structured workflows.\n
Ingestion points: User input through usage examples such as vendor comparisons and risk assessments, which are processed into workspace files like _workspace/00_input.md.\n
Boundary markers: No specific delimiters or safety instructions are present in the guides to isolate external content.\n
Capability inventory: File system write access to a local _workspace/ directory and routing capabilities to other specialized agent skills.\n
Sanitization: The instructions do not specify any sanitization or validation steps for incoming data.\n- [COMMAND_EXECUTION]: The skill's instructions involve managing project-specific files within a dedicated _workspace/ folder, including the creation of risk identification logs and monitoring plans. These operations are restricted to project management tasks and do not involve arbitrary or dangerous command execution.\n- [EXTERNAL_DOWNLOADS]: The skill includes references to trusted South Korean government domains, specifically ftc.go.kr (Fair Trade Commission) for procurement templates and hometax.go.kr (National Tax Service) for business status verification. These links are appropriate for the skill's procurement and compliance functions and do not involve the execution of external code.\n- [SAFE]: No malicious patterns, such as prompt injection, obfuscation, or persistence mechanisms, were identified. The skill correctly routes specialized legal and financial tasks to appropriate external tools, ensuring a controlled execution environment.

collab-vendor