sentry-advanced-troubleshooting

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • COMMAND_EXECUTION (MEDIUM): The skill is granted 'Bash(cmd:*)' permissions, providing unrestricted access to the shell environment. This is intended for 'sentry-cli' usage and dependency auditing (e.g., 'npm list'), but the excessive scope allows for potentially dangerous system commands.
  • PROMPT_INJECTION (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) because it instructs the agent to ingest and analyze external application logs and CLI outputs which may contain malicious instructions.
  • Ingestion points: Application logs, sentry-cli output, environment data.
  • Boundary markers: None identified in instructions to separate data from instructions.
  • Capability inventory: High-risk capabilities available including 'Bash', 'Write', and 'Edit'.
  • Sanitization: No sanitization or verification of external log data is performed before processing.
  • DATA_EXFILTRATION (LOW): The skill accesses the 'SENTRY_DSN' and performs network operations via Sentry SDK calls. The unrestricted 'Bash' tool could be misused to transmit sensitive environment variables or local files to external endpoints.
  • REMOTE_CODE_EXECUTION (LOW): The skill utilizes a pattern of generating local TypeScript files (e.g., 'diagnostic.ts') and executing them. While based on local templates, this dynamic execution of generated code increases the attack surface.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:41 PM