lisa-integration-test

Warn

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes arbitrary scripts from the target project (e.g., bun run typecheck, npm run lint, bun run test). These commands are defined in the project's package.json and represent a risk if the target project contains malicious configurations.
  • [DATA_EXFILTRATION]: The skill is designed to perform git push operations to remote repositories. While part of the intended workflow, this allows the agent to send local data to external servers automatically.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It reads project files (like .lisa-manifest) and processes error outputs from compilers and test runners to "diagnose" and "fix" code. A malicious project could use crafted error messages or file contents to influence the agent into performing unintended file modifications or command executions.
  • [COMMAND_EXECUTION]: The skill automatically modifies local files and stages them for commit (git add, git commit) based on its own analysis of build failures, which involves processing untrusted data from the project environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 16, 2026, 12:05 PM