The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses payroll data from external files, including employee names and table headers. This content is then presented back to the agent in the form of validation reports. A malicious file could contain prompt injection payloads designed to steer the agent's behavior once the results are read.
Ingestion points: scripts/run.mjs reads data from files provided via the --input argument.
Boundary markers: The tool splits data into cells using delimiters like tabs or commas but does not implement semantic boundaries or 'ignore' instructions for the agent regarding the parsed content.
Capability inventory: The skill is granted Bash and Read permissions in its configuration, creating an attack surface if the agent is influenced by injected text.
Sanitization: While numeric values are normalized, the skill does not sanitize or escape the string content of table cells before inclusion in reports.
[EXTERNAL_DOWNLOADS]: The SKILL.md documentation provides instructions for users to tell their AI assistant to install a supplementary package (@alipay/agent-payment) to facilitate a purchase. The package belongs to a well-known and reputable organization. This interaction is part of the skill's commercial workflow and is documented here for completeness.

payroll-check-free