NYC

grepai-troubleshooting

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Data Exposure & Exfiltration] (HIGH): The diagnostic command echo $OPENAI_API_KEY causes the agent to print the literal value of the secret API key into its output buffer, potentially exposing it to logging services or persisting it in the conversation history.\n- [Indirect Prompt Injection] (HIGH):\n
  • Ingestion points: The skill reads potentially untrusted local files including .grepai/config.yaml, ~/.claude/mcp.json, and .cursor/mcp.json.\n
  • Boundary markers: Absent; the skill does not use delimiters to wrap the output of cat commands, allowing data to be confused with instructions.\n
  • Capability inventory: The skill allows for destructive operations (rm -rf .grepai), environment variable modification (export), and local network interaction (curl).\n
  • Sanitization: Absent; the content of configuration files is processed as raw text without escaping or validation.\n- [Data Exposure & Exfiltration] (MEDIUM): The skill accesses ~/.claude/mcp.json and .cursor/mcp.json, which are configuration files for the host's AI tools located in the home directory, potentially exposing sensitive environment details.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 08:18 PM