cryptotokenkit

Warn

Audited by Gen Agent Trust Hub on Mar 28, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The documentation provides a shell command that uses sudo to execute a binary as the _securityagent user, which is a high-privilege operation. Evidence: 'sudo -u _securityagent /Applications/TokenHost.app/Contents/MacOS/TokenHost' in SKILL.md.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from hardware security tokens without explicit validation or boundary markers. 1. Ingestion points: TKSmartCard.send() and transmit() calls in SKILL.md and references/cryptotokenkit-patterns.md. 2. Boundary markers: Absent; there are no instructions to the agent to treat card data as untrusted or delimited. 3. Capability inventory: The skill has the capability to interact with the system keychain and perform cryptographic signatures/decryption. 4. Sanitization: While the code checks status words for communication success, it lacks logic to sanitize or validate the actual data payload returned from the token.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 28, 2026, 08:14 PM