market-research-writer

Pass

Audited by Gen Agent Trust Hub on Jun 25, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill utilizes curl to transmit local files and user data to the endpoint http://opencode-bridge:8000. This involves sending information from the agent's environment to a remote service for processing.
  • [COMMAND_EXECUTION]: The skill relies on shell command execution, specifically curl, to perform file uploads, manage tasks on the OpenCode bridge, and retrieve results.
  • [EXTERNAL_DOWNLOADS]: The skill is configured to download generated market reports and chart assets from the opencode-bridge server to the local /tmp/ directory.
  • [PROMPT_INJECTION]: The skill ingests untrusted user data through uploaded documents (Excel, CSV, PDF, etc.), which are then analyzed by an external execution engine. This creates a surface for indirect prompt injection.
  • Ingestion points: User-provided files uploaded via the http://opencode-bridge:8000/file/{user} endpoint (specifically Excel, CSV, PDF, Word, and PPT files).
  • Boundary markers: The skill advises the agent to reference files by name rather than content, but does not implement specific delimiters or "ignore" instructions for the data processed by the external tool.
  • Capability inventory: File system read/write access, network communication via curl, and task delegation to a remote execution environment.
  • Sanitization: No sanitization or content validation is performed on the files prior to ingestion or transmission.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 25, 2026, 02:21 PM
Security Audit — agent-trust-hub — market-research-writer