scenario-writer
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documentation instructs the execution of
bun run verify-tests, which performs file system operations and initiates the Vitest test runner to validate component implementations.- [REMOTE_CODE_EXECUTION]: The benchmarking workflow involves reading user-provided prompts, requesting an AI agent to generate code, writing that code to a temporary file, and executing tests against it. This process creates a path for the execution of untrusted code within the benchmarking environment.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by using user-supplied descriptions to drive the creation of code by a subagent. - Ingestion points: User descriptions provided to the scenario writer (referenced in SKILL.md).
- Boundary markers: None identified; user input is processed as instructions for the subagent.
- Capability inventory: The environment allows file system writes and shell execution (
bun,vitest). - Sanitization: The skill includes a static analysis validation prompt to ensure compliance with Svelte 5 syntax, but it does not perform security-specific sanitization or sandboxing of the generated code.
Audit Metadata