skills/cin12211/orca-q/graphify/Gen Agent Trust Hub

graphify

Fail

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs the 'graphifyy' package from PyPI. The name 'graphifyy' (with a double 'y') is highly suspicious as it is a one-character variation of the common 'graphify' package name, representing a potential typosquatting attack.
  • [REMOTE_CODE_EXECUTION]: Step 1 of the pipeline installs the unversioned 'graphifyy' package (and optional extras like 'graphifyy[gemini]' or 'graphifyy[video]') and immediately executes it. Running unverified, unversioned code from external sources is a critical security risk.
  • [COMMAND_EXECUTION]: The skill makes extensive use of the bash tool to execute complex multi-line Python scripts and system commands. Notably, it dispatches subagents with 'general-purpose' permissions, granting them access to the 'Bash' and 'Write' tools to perform data extraction on the local filesystem.
  • [PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection. It ingests arbitrary files (code, documents, images, videos) from the user's project and passes them to subagents without sanitization or boundary markers. Because these subagents have 'Bash' and 'Write' capabilities (Step 3B2 in SKILL.md), a malicious project file could execute arbitrary code on the host system.
  • Ingestion points: All files in the specified INPUT_PATH processed during Step 2 and 3.
  • Boundary markers: Absent; subagent prompts lack instructions to ignore embedded malicious content.
  • Capability inventory: Subagents have full 'Bash' and 'Write' access via the 'general-purpose' agent type.
  • Sanitization: None; raw file content is directly interpolated into extraction prompts.
  • [DATA_EXFILTRATION]: The skill supports cloning external GitHub repositories to the local environment and can push extracted graph data to remote Neo4j instances via the 'neo4j-push' flag.
  • [CREDENTIALS_UNSAFE]: The skill instructions prompt the user to set sensitive environment variables like 'GEMINI_API_KEY' or 'GOOGLE_API_KEY' and accepts Neo4j credentials, including passwords, via command-line arguments.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 30, 2026, 04:47 PM
Security Audit — agent-trust-hub — graphify