adaptive-decision-interviews
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill directs the agent to read sensitive data sources, specifically application logs and source code, to gather context for interviews ("Inspect the available source pack: tickets, research, prior decisions, code, logs, and existing artefacts"). This creates exposure to potentially sensitive information stored in those files. The skill provides a specific mitigation by instructing the agent to redact credentials and secrets discovered during this process using a specific placeholder format ([REDACTED]).- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external artifacts such as tickets and log files, which serves as an ingestion point for potential indirect prompt injection attacks. A malicious entry in a log file could attempt to influence the agent's interpretation or interview flow. The skill mitigates this risk by requiring the agent to record human answers verbatim, separate machine interpretation from raw evidence, and maintain a human-in-the-loop requirement for all authorized decision artifacts.
- Ingestion points: Source pack artifacts including tickets, research documents, logs, and code files.
- Boundary markers: Instructions mandate the use of verbatim block quotes for human answers and explicit separation of interpretations from evidence.
- Capability inventory: The skill utilizes file system read capabilities for source packs and file system write capabilities for generating interview records.
- Sanitization: Includes explicit instructions for the agent to filter and redact credentials or secrets found in the data.
Audit Metadata