candidate-screening
Warn
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands, specifically
npm run screenandcat, to interact with a local screening tool and its output files. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted candidate data from resumes and profiles. A malicious applicant could include instructions to manipulate the scoring or behavior of the agent.
- Ingestion points: Job applications, resumes, and LinkedIn profiles.
- Boundary markers: Absent; there are no instructions to the agent to treat candidate data as untrusted or to use delimiters.
- Capability inventory: Shell command execution (
npm), file system access (cat), and network transmission of summaries (--send-slack). - Sanitization: No sanitization or validation of the input candidate data is performed before processing.
- [DATA_EXFILTRATION]: The skill manages sensitive Personal Identifiable Information (PII). It facilitates the extraction of candidate details from an ATS and provides mechanisms to send summaries to external platforms like Slack.
Audit Metadata