36kr-hotlist
Fail
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: HIGHINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests article titles, authors, and summaries from the 36kr API and displays them to the user without delimiters or content sanitization. | Ingestion points: SKILL.md and fetch scripts in the scripts/ folder. | Boundary markers: The instructions do not define boundary markers to separate article content from agent instructions. | Capability inventory: The article content is for display only and is not passed to dangerous tools like file-writers or command executors. | Sanitization: No validation or filtering is performed on the ingested data before presentation.
- [COMMAND_EXECUTION]: Documentation provides shell examples that pipe remote API content to the Python interpreter. | Evidence: curl commands in SKILL.md and examples.md pipe directly to 'python3 -m json.tool'. | While 'json.tool' is a safe built-in module for formatting, the pattern of piping network results to an interpreter is a sensitive practice.
- [EXTERNAL_DOWNLOADS]: The skill performs network requests to retrieve JSON data from 'openclaw.36krcdn.com'. | This domain is a legitimate infrastructure resource belonging to the skill author (36kr-com).
Recommendations
- HIGH: Downloads and executes remote code from: https://openclaw.36krcdn.com/media/hotlist/$(date - DO NOT USE without thorough review
Audit Metadata