category-selection

Pass

Audited by Gen Agent Trust Hub on Mar 20, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches market data, product lists, and statistics from the external domain mcp.sorftime.com through multiple Python scripts including workflow.py and analyze_category.py.
  • [COMMAND_EXECUTION]: The scripts/workflow.py script uses subprocess.run to execute curl for API calls. Additionally, the skill's instructions in SKILL.md guide the agent to run local Python scripts to process data and generate reports.
  • [DATA_EXFILTRATION]: The skill retrieves a user's Sorftime API key from environment variables or a local .mcp.json file and transmits it to https://mcp.sorftime.com for authentication. This is required for functionality but involves sending credentials to a third-party service.
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting and processing untrusted data from the Sorftime API. Evidence Chain: 1. Ingestion points: Data is fetched from the external API in scripts/workflow.py and processed in scripts/analyze_category.py. 2. Boundary markers: No explicit boundary markers or instructions to ignore embedded prompts are present in the report generation templates found in assets/report_template.md or assets/dashboard_template.html. 3. Capability inventory: The skill can execute shell commands via curl and write files to the local file system. 4. Sanitization: While the scripts include logic to clean control characters and fix encoding (Mojibake), they do not sanitize text for potential LLM instructions embedded in the external product data.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 20, 2026, 10:55 AM