feishu-projects

Warn

Audited by Gen Agent Trust Hub on Apr 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The references/attachments.md file explicitly directs the agent to execute shell commands using curl for file operations: curl -X POST "<returned_url>" ... --data-binary @<local_file>. This instruction bypasses structured tool interfaces in favor of direct shell interaction. \n- [DATA_EXFILTRATION]: The defined workflow for uploading attachments involves reading the binary content of local files and transmitting it to external URLs. While intended for Feishu project management, this capability represents a potential path for exfiltrating sensitive local data if the destination URL or file path is manipulated. \n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to download content from remote URLs to the local system using command-line tools. \n- [PROMPT_INJECTION]: The skill processes untrusted data from Feishu work items and local configuration files, which could lead to indirect prompt injection. \n
  • Ingestion points: Feishu API outputs (work item fields, descriptions, and comments) and the .agents/constant.md configuration file. \n
  • Boundary markers: The instructions do not specify any delimiters or safety markers to differentiate between instructions and data when processing external content. \n
  • Capability inventory: File system access (reading configuration, reading files for upload), network access via curl, and a suite of tools for project modification. \n
  • Sanitization: The skill does not include instructions for validating or sanitizing external data before it is rendered or used in query construction.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 23, 2026, 03:31 AM
Security Audit — agent-trust-hub — feishu-projects