arkweb-architect
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill manages the local development environment and source code through shell and git operations.
- Initializes project structures using
mkdir -pto organize design and analysis artifacts. - Utilizes
git clone --depth=1to download specific repository versions for technical assessment in temporary local directories. - Performs code lifecycle management tasks including
git difffor change verification, followed bygit commitandgit pushlogic for submission. - [EXTERNAL_DOWNLOADS]: Fetches technical metadata and source code from established documentation and hosting services.
- Communicates with the GitCode API to retrieve knowledge base evidence from OpenHarmony and Chromium related repositories.
- Accesses architectural documentation and source indexing from
deepwiki.comto supplement subagent analysis tasks. - [PROMPT_INJECTION]: The skill ingests and processes various forms of external data, creating an attack surface for indirect instruction influence.
- Processes untrusted content originating from user-supplied requirements, third-party documentation APIs, and code from remote repositories.
- This ingested content is interpolated directly into the task descriptions for spawned subagents.
- The current design lacks explicit boundary markers or instruction-bypass warnings to delineate user/external data from the agent's core instructions.
Audit Metadata