understand-chat

Warn

Audited by Gen Agent Trust Hub on Apr 18, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill's instructions guide the agent to use user-provided search terms ('$ARGUMENTS') directly within shell-based grep commands. This creates a risk of command injection if the agent's execution environment does not provide robust isolation or sanitization for shell arguments.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from a local JSON file ('.understand-anything/knowledge-graph.json') to formulate its responses. This represents an indirect prompt injection vulnerability where an attacker could place malicious instructions in node summaries or tags to hijack the agent's logic.
  • Ingestion points: .understand-anything/knowledge-graph.json
  • Boundary markers: Absent; there are no instructions to delimit the data or ignore embedded commands.
  • Capability inventory: File reading, shell command execution (via grep).
  • Sanitization: Absent; the skill does not suggest any validation or filtering of the graph content before it enters the context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 18, 2026, 12:08 PM
Security Audit — agent-trust-hub — understand-chat