Composio Connect

Fail

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Fetches and executes an installation script directly from the vendor's official domain (https://composio.dev/install) using a piped-to-shell pattern.
  • [COMMAND_EXECUTION]: Relies on the execution of the composio CLI tool and various shell utilities (jq, grep, awk, sed) for data manipulation and system interaction.
  • [DATA_EXFILTRATION]: Accesses and processes data from external applications like Slack, GitHub, and Gmail. While it includes warnings against hardcoding credentials, the skill involves handling potentially sensitive project and user metadata.
  • [PROMPT_INJECTION]: Presents a surface for indirect prompt injection as it processes untrusted data from over 1000 external applications and has the capability to execute commands based on that data.
  • Ingestion points: Data retrieved from external apps via the composio tools execute command (SKILL.md).
  • Boundary markers: Absent for external tool outputs.
  • Capability inventory: CLI execution (composio), and instructions to offload work to inline bash or python scripts (SKILL.md).
  • Sanitization: Suggests parsing JSON with jq but lacks explicit sanitization of text content from external sources.
Recommendations
  • HIGH: Downloads and executes remote code from: https://composio.dev/install - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 6, 2026, 02:24 AM