mcp-builder

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [Data Exposure & Exfiltration] (LOW): The skill provides a template for a 'read_file' tool ('open(path).read()') that allows reading arbitrary files from the system without path validation or sandboxing.
  • [Indirect Prompt Injection] (LOW): The skill defines several ingestion points for untrusted data (tool arguments) without implementing robust boundary markers or sanitization. 1. Ingestion points: 'query_db(sql)' and 'read_file(path)' in SKILL.md. 2. Boundary markers: Absent in provided templates. 3. Capability inventory: Includes file system access ('open'), network requests ('httpx'), and database interaction ('sqlite3'). 4. Sanitization: Minimal; the SQL example only checks for the 'SELECT' keyword, which is insufficient.
  • [External Downloads] (SAFE): The skill uses standard libraries and a tool from a trusted source (@anthropics).
  • [False Positive Note] (SAFE): Automated scanner alerts regarding 'request.params.name' are confirmed false positives as it is a code property reference.
Recommendations
  • Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 17, 2026, 05:50 PM