record

Fail

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill constructs shell commands in Phases 2, 4, and 6 by interpolating session-derived variables such as <last-action>, <next-action>, <alias-key>, and <full-meaning> directly into double-quoted Bash strings. This pattern is vulnerable to command injection via shell subshell expansion (e.g., $(...) or backticks) if the variables contain malicious sequences.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted session data and utilizing it in shell commands.
  • Ingestion points: Session state data is gathered from tool history, project files, and user dialogue in Phase 1 and 6.
  • Boundary markers: No delimiters or instructions to ignore embedded commands are present during variable interpolation.
  • Capability inventory: The skill uses the Bash tool to execute logic via the memplan-cli.js script and the rm command.
  • Sanitization: No evidence of escaping or validation is present for the session-derived variables before they are passed to the shell.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 19, 2026, 05:23 AM
Security Audit — agent-trust-hub — record