lov-media-crawler

Warn

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill facilitates the execution of scripts that require administrative privileges to modify system-level configurations.
  • Evidence: scripts/wxclient.sh uses sudo to run the wx_video_download binary, which is designed to install a root certificate and configure a system-wide proxy.
  • [EXTERNAL_DOWNLOADS]: Setup routines fetch code and pre-compiled binaries from non-whitelisted third-party GitHub repositories.
  • Evidence: scripts/media_crawler.py downloads a macOS binary from github.com/ltaoo/wx_channels_download and clones a code repository from github.com/NanmiCoder/MediaCrawler.
  • [DYNAMIC_EXECUTION]: The tool assembles and executes Python code dynamically to inject configuration into external crawler components at runtime.
  • Evidence: scripts/media_crawler.py constructs a Python script string and executes it via uv run python -c to initiate the MediaCrawler adapter with specific configuration overrides.
  • [COMMAND_EXECUTION]: Extensive use of shell commands to manage system credentials, external dependencies, and file system operations.
  • Evidence: The implementation uses subprocess.run to interact with the macOS Keychain (security tool), git, uv, and various download utilities like aria2c and curl.
  • [DATA_EXFILTRATION]: Provides functionality to transmit data to an external, third-party resolution service.
  • Evidence: The skill defines and utilizes https://sph.litao.workers.dev/api/fetch_video_profile as a remote resolver when the --allow-public-resolver flag is enabled.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from external social media platforms which directly influences file system write paths.
  • Ingestion points: The feed_metadata and wxclient_metadata functions in scripts/media_crawler.py extract titles and descriptions from social media API responses.
  • Boundary markers: Absent; there are no instructions to the agent to ignore embedded commands in the metadata.
  • Capability inventory: The skill has the capability to write to the file system via the destination_path and run_downloader functions.
  • Sanitization: Employs a safe_filename regex-based filter to strip restricted characters from strings before using them in file paths.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 17, 2026, 03:41 PM