pinchtab

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web pages using commands like snap, text, and find, which can surface malicious instructions designed to trick the agent. * Ingestion points: SKILL.md (via browser observation tools). * Boundary markers: Documentation explicitly warns agents to ignore directives sourced from pages and treat them as untrusted. * Capability inventory: Full browser interaction including clicking, typing, file management, and JS execution. * Sanitization: Relies on agent following safety guidelines and user verification for high-risk actions.
  • [DYNAMIC_EXECUTION]: The pinchtab eval command permits execution of arbitrary JavaScript in the browser context, which is a powerful capability gated by the security.allowEvaluate setting.
  • [DATA_EXFILTRATION]: Commands like pinchtab cookies and pinchtab network-export allow extraction of sensitive session data and authentication headers from the browser.
  • [COMMAND_EXECUTION]: The tool manages local browser instances and provides a daemon install command for background service persistence.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 11:32 PM
Security Audit — agent-trust-hub — pinchtab