cangjie-dev

Pass

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: LOW
Full Analysis
  • Prompt Injection (SAFE): The skill instructions establish a helpful persona as a Cangjie language expert and do not contain any instructions intended to bypass safety filters or override system constraints.\n- Data Exposure & Exfiltration (SAFE): No hardcoded credentials or sensitive file path access was detected. Standard database connection examples utilize benign local placeholders.\n- Obfuscation (SAFE): Analysis across the documentation set found no multi-layer obfuscation, hidden unicode characters, or homoglyphs used to mask malicious intent.\n- Unverifiable Dependencies & RCE (SAFE): The skill contains no executable scripts. Documentation for toolchain installation refers to official Huawei and standard OpenSSL sources, with no automated remote execution patterns like piped curl-to-bash commands.\n- Indirect Prompt Injection (INFO): The skill identifies a data ingestion surface, as it is designed to process user-provided Cangjie source code (.cj) and configuration files (cjpm.toml). However, the agent's capabilities are limited to reasoning and generating code snippets, with no high-privilege tool access.\n
  • Ingestion points: .cj and cjpm.toml files.\n
  • Boundary markers: Absent in instructions.\n
  • Capability inventory: None (The skill lacks automated execution or data transmission scripts).\n
  • Sanitization: Absent.\n- Dynamic Execution (SAFE): While the skill documents Cangjie's reflection and macro features, the documentation content itself does not execute dynamic code or employ unsafe deserialization.
Audit Metadata
Risk Level
LOW
Analyzed
Feb 16, 2026, 01:36 AM