devils-advocate

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local shell script review-state-log.sh to record review outcomes. This command uses several parameters, including --notes and --paper, which are populated based on the analysis of external input.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted research papers and uses the resulting critiques as arguments for a shell command.- [INDIRECT_PROMPT_INJECTION]: The following evidence chain was identified:
  • Ingestion points: Research papers or argument descriptions provided by the user are the primary source of untrusted data (SKILL.md).
  • Boundary markers: No specific delimiters or instructions to ignore embedded commands within the analyzed documents are present.
  • Capability inventory: The skill performs directory creation (mkdir), file writing, and shell script execution (bash) (SKILL.md).
  • Sanitization: The skill lacks instructions to sanitize or escape special characters in the text used to populate the --notes or --paper arguments, creating a risk of command injection if the input content is malicious.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 06:16 AM
Security Audit — agent-trust-hub — devils-advocate