The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data—such as business line names, revenue figures, and resource consumption—to generate formatted reports and tables. While this creates a potential surface for indirect prompt injection, the skill lacks exploitable capabilities (such as file system writes, shell command execution, or external network requests), and the risk is mitigated by its structured analytical workflow.
Ingestion points: Phase 0.2 collects user-supplied business line names and performance metrics.
Boundary markers: The skill uses a multi-phase workflow with specific Markdown templates to isolate analysis steps.
Capability inventory: Limited to data analysis and text output; no subprocess calls or network exfiltration tools are used.
Sanitization: The skill includes a specific 'Privacy Mode' to pseudonymize sensitive identifiers (names, IDs, domains) into generic aliases before processing.
[SAFE]: The skill references an external documentation and upgrade link to the author's GitHub repository (github.com/iamzifei/aikey). This is a legitimate vendor-owned resource used for user support.
[SAFE]: The skill implements a 'pseudonym' privacy mode that mandates the replacement of real business names, domains, and IDs with aliases. This design choice follows security best practices for handling potentially sensitive business information in an AI context.

aikey-portfolio