test-and-fix
Warn
Audited by Gen Agent Trust Hub on Jun 12, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructs the agent to automatically search for and access sensitive files such as
.credentials.jsonandCLAUDE.mdto extract authentication credentials for automated testing purposes. - [COMMAND_EXECUTION]: The skill involves the dynamic generation and execution of Python scripts (utilizing Playwright) to perform functional and stress tests. It also grants the agent the capability to directly modify the application's source code to fix detected bugs, which involves high-privilege file system operations.
- [PROMPT_INJECTION]: The skill exhibits vulnerability to indirect prompt injection. It ingests untrusted data from external sources (web page HTML, console logs, and API responses) and uses this data to inform its bug-fixing logic and agent reviewers.
- Ingestion points: Target web application pages, console errors, network failure logs, and project route files.
- Boundary markers: None. There are no instructions provided to distinguish between the content being tested and the agent's internal control instructions.
- Capability inventory: Execution of Python/Playwright scripts, shell command execution (curl), modification of source code files, and spawning of parallel background agents.
- Sanitization: None. The skill does not validate or sanitize the content extracted from the web application before using it to generate fixes or reviewer reports.
Audit Metadata