zlib-download

Warn

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The 'scripts/setup.sh' script downloads a pre-compiled 'annas-mcp' binary from a third-party GitHub repository ('iosifache/annas-mcp') and installs it locally. This binary is then executed by the AI agent via the 'scripts/book.py' wrapper to perform book searches and downloads.
  • [EXTERNAL_DOWNLOADS]: During setup, the skill fetches an executable binary and Python dependencies from external, unverified sources.
  • [COMMAND_EXECUTION]: The skill uses 'subprocess.run' to call the externally downloaded 'annas-mcp' binary. While arguments are passed as a list to prevent shell injection, the skill's operation depends on the integrity of this third-party executable.
  • [DATA_EXFILTRATION]: The skill prompts the user to store Z-Library credentials in a local '.env' file, which are then read and sent to the '1lib.sk' domain for authentication. This involves transmitting sensitive plaintext credentials to a shadow library mirror.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions.
  • Ingestion points: Book metadata (titles, authors) is fetched from Z-Library and Anna's Archive APIs in 'scripts/book.py'.
  • Boundary markers: Absent. The agent is instructed to present the raw metadata in a table format.
  • Capability inventory: The skill can execute shell commands (via subprocess), write files to the local system, and make network requests.
  • Sanitization: Filenames are sanitized for Z-Library downloads, but general metadata displayed to the user is not filtered.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 17, 2026, 04:24 AM
Security Audit — agent-trust-hub — zlib-download