error-recovery

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • PROMPT_INJECTION (HIGH): The skill is highly vulnerable to Indirect Prompt Injection (Category 8).\n
  • Ingestion points: It reads external content from error-log.txt, command outputs, and stack traces.\n
  • Boundary markers: There are no delimiters or 'ignore' instructions wrapping the untrusted error data.\n
  • Capability inventory: The skill can execute git reset --hard, rm -rf, and ./scripts/init.sh.\n
  • Sanitization: No sanitization is performed on error content before it is processed or posted to GitHub via gh issue comment. An attacker could craft a 'failure' containing instructions to influence the agent's next steps.\n- COMMAND_EXECUTION (HIGH): The skill executes potentially destructive commands and local scripts.\n
  • Evidence: The use of git reset --hard and rm -rf node_modules can be used to cause data loss or remove security controls.\n
  • Evidence: The execution of ./scripts/init.sh during recovery is a significant risk; if a previous step or a malicious actor modified this script before the 'failure', the agent would execute the compromised code.\n- DATA_EXFILTRATION (MEDIUM): The skill risks exposing sensitive environment data.\n
  • Evidence: It explicitly runs env | grep -E "NODE|NPM|PATH" and directs the output to logs or GitHub comments. Secrets are frequently stored in environment variables, and broad grep patterns may capture them.\n- EXTERNAL_DOWNLOADS (MEDIUM): The skill executes code from external sources without version pinning.\n
  • Evidence: It uses pnpm dlx madge to fetch and run a package from the npm registry at runtime, which is susceptible to supply chain attacks.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 05:26 AM