zlib-download-skill

Fail

Audited by Gen Agent Trust Hub on Feb 20, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • EXTERNAL_DOWNLOADS (HIGH): The setup.sh script downloads a pre-compiled binary (annas-mcp) from an untrusted GitHub repository (iosifache/annas-mcp) that is not on the permitted list of trusted sources.
  • REMOTE_CODE_EXECUTION (HIGH): The skill downloads an external binary, sets execution permissions using chmod +x, and executes it via the book.py wrapper. This provides a direct path for arbitrary code execution if the remote binary or repository is compromised.
  • CREDENTIALS_UNSAFE (HIGH): The skill facilitates the storage of plaintext Z-Library and Anna's Archive credentials in a local configuration file (~/.claude/book-tools/.env). Exposure of this file would lead to a full credential compromise.
  • COMMAND_EXECUTION (MEDIUM): The setup.sh script performs shell-level operations including downloading, extracting, and installing software to the user's local binary directory.
  • INDIRECT_PROMPT_INJECTION (LOW): The skill ingests untrusted book metadata from external search APIs and presents it to the agent, creating a vulnerability surface. * Ingestion points: scripts/book.py search results. * Boundary markers: Absent; no delimiters are used to separate untrusted book data from agent instructions. * Capability inventory: The skill can execute subprocesses via scripts/book.py download. * Sanitization: Absent; external data is presented directly to the agent without filtering or escaping.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 20, 2026, 04:51 AM