wizard

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill generates bash scripts designed to perform system actions, including opening URLs via platform-specific commands (such as xdg-open or explorer.exe) and managing repository configurations using the GitHub CLI (gh). These operations are intended to be executed by the human user in a guided workflow.
  • [DATA_EXFILTRATION]: The skill's instructions require the agent to read local files, including .env and GitHub workflow definitions, to identify required secrets. The generated scripts are designed to capture sensitive values (e.g., API keys) from the user and transmit them to external services like GitHub Secrets or save them to local environment files. While these are documented project setup steps, they involve the handling of sensitive credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes content from various repository files, such as READMEs and configuration files, to determine the necessary setup steps. This ingestion of potentially untrusted data represents a vulnerability surface where malicious content within a repository could attempt to influence the logic of the generated wizard.
  • Ingestion points: The agent is instructed to read .env, README, docker-compose, and .github/workflows/* files (SKILL.md).
  • Boundary markers: No explicit delimiters or warnings to ignore instructions within the read files are provided in the skill logic.
  • Capability inventory: Generated scripts possess capabilities to write files (write_env), open web browsers (open_url), and interact with GitHub's API (set_secret, set_var).
  • Sanitization: There is no evidence of sanitization or filtering of the content read from the repository files before it is used to scope the procedure.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 09:28 AM