bioresearcher-python-setup-uv
Fail
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill employs absolute directives to override the agent's normal operational logic and prevent modification of provided commands.
- Evidence: 'ABSOLUTE RULE: Follow steps below EXACTLY AS IS. Do NOT skip/modify steps (nor detailed subtasks in each step) nor assume anything based on user platform information.' in
SKILL.md. - [REMOTE_CODE_EXECUTION]: The skill facilitates the execution of remote scripts by piping downloaded content directly into shell and PowerShell interpreters.
- Evidence: Execution patterns like
curl -LsSf UV_INSTALLER_URL | ... shandInvoke-RestMethod UV_INSTALLER_URL | Invoke-Expressionare used to install the uv tool. - Source: While an official source is provided, the skill also promotes a personal Gitee repository (
https://gitee.com/wangnov/uv-custom) as a primary mirror, which is an unverified source for security-critical scripts. - [INDIRECT_PROMPT_INJECTION]: The skill modifies persistent agent instruction files to influence future behavior without providing boundary markers or sanitization.
- Ingestion points:
AGENTS.md(orCLAUDE.md). - Boundary markers: Absent.
- Capability inventory:
Bashtool (curl, sh, uv), file system write access. - Sanitization: Absent; instructions are appended verbatim to the root configuration file.
- [COMMAND_EXECUTION]: The skill uses the shell to perform file system operations and binary management.
- Evidence: Use of
ln -sf,mklink,copy /Y, and./uv pip installcommands inSKILL.md.
Recommendations
- AI detected serious security threats
Audit Metadata