gangtise-copilot
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches official Gangtise skill bundles from a Huawei Cloud OBS bucket (gts-download.obs.myhuaweicloud.com). This is a well-known cloud storage service used by the vendor to distribute tools.
- [COMMAND_EXECUTION]: Uses shell scripts to automate the installation and credential setup process. The scripts follow security best practices by applying restrictive permissions (chmod 600) to sensitive configuration files like authorization.json.
- [PROMPT_INJECTION]: The skill suite possesses an attack surface for indirect prompt injection through its data ingestion components.
- Ingestion points: Untrusted data such as research reports, meeting summaries, and WeChat chat logs enter the agent context via the kb-client and wechat-summary tools documented in references/skill_registry.md.
- Boundary markers: The wrapper logic lacks explicit instruction delimiters or markers to differentiate between system instructions and processed data.
- Capability inventory: The installed skills are capable of executing network requests and reading local files to perform research tasks.
- Sanitization: There is no evidence of sanitization or filtering of external content before it is interpolated into agent prompts.
Audit Metadata