e2e-flow-verifier
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data and variables to drive browser automation, which introduces a surface for indirect instructions.
- Ingestion points: User-defined flow parameters (e.g.,
base_url,test_user,expected_text) are interpolated into JSON batch files and shell commands withinSKILL.md. - Boundary markers: None are present in the provided templates to isolate user-supplied data from the command execution flow.
- Capability inventory: The skill executes the
vibiumbinary for browser interaction, runs local Node.js scripts (batch.js,assert.js), and performs file writes to/tmpandtest-results/. - Sanitization: There is no evidence of sanitization or validation for the interpolated variables within the skill instructions.
- [COMMAND_EXECUTION]: The skill provides instructions to execute local binaries and scripts to perform its primary function.
- Execution details: It invokes the
vibiumCLI for recording browser sessions (vibium record) and restoring session state (vibium storage). It also utilizesnodeto execute scripts located in.claude/skills/qe-browser/for batch processing and state assertions.
Audit Metadata