The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses highly sensitive HR information, including Personal Identifiable Information (PII) such as candidate names, contact details, employment history, and resume attachments from ServiceNow tables like sn_hr_tm_candidate and sn_hr_core_profile.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the processing of untrusted candidate-provided data.
Ingestion points: Data is ingested from the resume_text and resume_content fields of the candidate table, as well as file content from the sys_attachment table.
Boundary markers: The instructions do not define clear boundary markers or delimiters to separate the untrusted resume content from the agent's instructions, increasing the risk that the agent may follow instructions embedded within a resume.
Capability inventory: The skill has the capability to modify records via SN-Update-Record and access local system resources through the Bash tool.
Sanitization: No sanitization or validation of the ingested resume text is specified in the procedure, relying on the LLM's internal filters to distinguish between data and instructions.

resume-skill-extraction