consultant

Fail

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill's documentation (SKILL.md) provides a command to install the 'uv' package manager via curl -LsSf https://astral.sh/uv/install.sh | sh. This is a standard installation method for a well-known developer tool from its official domain.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the analysis of arbitrary project files, creating a surface for indirect prompt injection where malicious instructions inside files could influence the consultant LLM's behavior.
  • Ingestion points: The --file argument in scripts/consultant_cli.py allows the user to specify file paths which are then read by the FileHandler.process_files method in scripts/file_handler.py.
  • Boundary markers: The skill uses delimiters to separate user prompts from file content. In scripts/file_handler.py, the build_prompt_with_references function wraps file contents with equals-sign headers (e.g., "=" * 80) and specific section titles like ## Reference Files.
  • Capability inventory: The skill has the capability to perform network requests (via litellm and requests in scripts/litellm_client.py), read files from the disk (in scripts/file_handler.py), and write session metadata and output to the local filesystem (in scripts/session_manager.py).
  • Sanitization: There is no specific sanitization or escaping applied to the content of the files before they are interpolated into the final prompt string.
  • [COMMAND_EXECUTION]: The skill utilizes uvx (a tool-runner from Astral) to execute its Python scripts and manage dependencies dynamically. Additionally, scripts/session_manager.py uses the multiprocessing.Process module to run consultation tasks in the background.
  • [CREDENTIALS_UNSAFE]: The skill handles sensitive API keys (e.g., LITELLM_API_KEY, OPENAI_API_KEY) through command-line arguments and environment variables. While this is standard for LLM tools, providing keys via CLI flags can expose them in process lists or shell history. The skill correctly recommends using environment variables as a safer alternative in its documentation.
Recommendations
  • HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 15, 2026, 08:12 AM