linkfox-temu-returns-refunds-global

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMPERSISTENCEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PERSISTENCE]: The skill provides instructions in references/onboarding.md for the agent to modify shell configuration files such as ~/.bashrc and ~/.zshrc, as well as the Windows registry via the setx command, to persist the LINKFOX_AGENT_API_KEY environment variable across sessions.
  • [DATA_EXFILTRATION]: The skill handles sensitive authentication secrets by storing Temu access tokens in a local plaintext JSON file located at ~/.linkfox/temu-access-tokens.json, as implemented in scripts/_temu_token_store.py. Furthermore, scripts/_temu_common.py defines a logging mechanism that writes complete API response bodies to the linkfox/ directory within the user's workspace, which could lead to the exposure of sensitive business or transactional data.
  • [COMMAND_EXECUTION]: Both the documentation in references/onboarding.md and the scripts/onboarding.py script include commands for environment setup, including pip installations for third-party libraries and shell commands for persistent environment variable configuration.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes external data from Temu API responses via scripts/temu_proxy.py and other service scripts without implementing explicit boundary delimiters or instructions to ignore potential malicious content within those responses.
  • Ingestion points: API response bodies received from the LinkFox gateway and processed in scripts/_temu_common.py and scripts/temu_proxy.py.
  • Boundary markers: None identified in the prompt logic.
  • Capability inventory: Network operations via urllib.request and requests, and file system write operations for logging and token storage.
  • Sanitization: The skill employs standard JSON parsing but lacks specific filtering to prevent adversarial instructions from external data from influencing agent behavior.
  • [EXTERNAL_DOWNLOADS]: The onboarding process recommends the installation of several external Python packages, specifically requests, qrcode, and pillow, using the pip package manager.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 07:12 AM